Idioms Processing, Structure and Interpretation - Cristina Cacciari

IDIOMS: Processing,Structure, and Interpretation

IDIOMS: Processing,Structure, and Interpretation

Edited by

Cristina Cacciari

University of Bologna, Italy

Patrizia Tabossi

University of Ferrara, Italy

@w LAWRENCE ERLBAUM ASSOCIATES, PUBLISHERS1993 Hillsdale, New Jersey Hove and London

Copyright © 1993 by Lawrence Erlbaum Associates, Inc.All rights reserved. No part of this book may be reproduced inany form, by photostat, microform, retrieval system, or any othermeans, without the prior written permission of the publisher.

Lawrence Erlbaum Associates, Inc., Publishers365 BroadwayHillsdale, New Jersey 07642

Library of Congress Cataloging-in-Publication Data

Idioms: processing, structure, and interpretation / edited by CristinaCacciari and Patrizia Tabossi.

p. cm.Includes bibliographical references and index.ISBN 0-8058-1038-21. Idioms. 2. Psycholinguistics. 3. Language acquisition.

4. Semantics. I. Cacciari, C. (Cristina) 11. Tabossi, Patrizia.P301.5.13413 199340t'.9-dc2O 92-28580

CIP

Books published by Lawrence Erlbaum Associates are printed on acid-free paper, and theirbindings are chosen for strength and durability.

Printed in the United States of America

10 9 8 7 6 5 4 3 2 1

CONTENTS

ForewordP. N. Johnson-Laird

Preface

List of Contributors

PART I. IDIOM INTERPRETATION AND THELITERAL FIGURATIVE DISTINCTION

1. Idiom Meanings and Allusional ContentSam Glucksberg

2. The Place of Idioms in a Literal and Metaphorical WorldCristina Cacciari

3. Why Idioms Are Not Dead MetaphorsRaymond W. Gibbs, Jr.

4. The Comprehension and Semantic Interpretation of IdiomsGiovanni B. Flores d'Arcais

vii

xi

xvii

1

3

27

57

79

V

Vi CONTENTS

PART 11. ACQUISITION AND PROCESSINGOF IDIOMS 99

5. The Acquisition of Idioms and the Development of FigurativeCompetence 101

M. Chiara Levorato

6. Ill-Formedness and Transformability in Portuguese Idioms 129Teresa Botelho da Silva and Anne Cutler

7. The Activation of Idiomatic Meaning in Spoken LanguageComprehension 145

Patrizia Tabossi and Francesco Zardon

8. The Comprehension of Ambiguous Idioms in Context 163Lucia Colombo

9. Syntactic and Semantic Processing During IdiomComprehension: Neurolinguistic and PsycholinguisticDissociations 201

Robert R. Peterson and Curt Burgess

PART III. MEANING AND STRUCTURE 227

10. Building Castles in the Air: Some Computational andTheoretical Issues in Idiom Comprehension 229

Oliviero Stock, Jon Slack, and Andrew Ortony

11. The Recognition and Interpretation of Idioms 249Stephen G. Pulman

12. The Determiner in English Idioms 271Christiane Fellbaum

13. Idiomaticity as a Reaction to L 'Arbitraire du Signe in theUniversal Process of Semeio-Genesis 297

Adam Makkai

Author Index 325Subject Index 331

Foreword

If natural language had been designed by a logician, idioms would not exist.They are a feature of discourse that frustrates any simple logical account of howthe meanings of utterances depend on the meanings of their parts and on thesyntactic relation among those parts. Idioms are transparent to native speakers,but a course of perplexity to those who are acquiring a second language. Ifsomeone tells me that Mrs. Thatcher has become the Queen of Scotland, I amlikely to say: "That's a tall story. Pull the other one!" As anyone struggling tolearn English will aver, stories cannot be tall-they have no height, and so theexpression violates a restriction on the normal sense of the word. Similarly, topull something is a physical event, and "one" is a pronoun that normally harksback to something that has occurred in the discourse. But I am certainly notinviting you to make any physical action. "Tall" has an idiomatic sense ofimprobable or farfetched. "Pull the other one" has an idiomatic sense that relatesto another idiom: "to pull someone's leg," which means to joke or to teasesomeone. "To pull the other one" stands in for "to pull my other leg," and isaccordingly an ironic invitation to tell me another joke.

A simple litmus, though not an infallible one, for whether a sense is idiomaticis to consider its expression in another language. An Italian speaker, for example,does not say, "Mi stai tirando la gamba" [literally, you are pulling my leg] toexpress the idea of a joke, but rather: "Mi prendi in giro." This expression ispurely idiomatic. It does not have a literal meaning, though it clearly relates to:"Fammi fare un giro," which means you are taking me on a tour. Hence, someexpressions have both a literal and an idiomatic meaning, such as "It's betterthan a poke in the eye with a burnt stick," and others have only an idiomaticinterpretation, such as "You're giving me the run around." All this linguistic

vii

Viii FOREWORD

filigree is interesting no doubt, but does it really call for psycholinguistic studies,and for the book of scholarly essays that you presently hold in your hand? Theanswer is: yes. And to justify this claim, I will briefly explain why idioms areimportant, why their study is so central to current psychological investigations oflanguage, and why indeed all students of language as a mental phenomenon arelikely to benefit from reading the chapters in this book. Then I will pass you on tothe editors of the volume to introduce its contents to you.

The logical approach to language relegates idioms to the sidelines. Speakersuse idiomatic expressions, on this account, as though they were words or phrasesthat have become frozen into a single form with a special meaning. Somewherein the mind these expressions are stored as exceptions, much as they might belisted at the end of a dictionary. Listeners first try to make a literal interpretationof an utterance. The literal meaning can be constructed compositionally, that is,it can be composed from the meanings of its parts according to the syntacticrelations amongst them. Words have meanings; each syntactic rule in the gram-mar has a corresponding semantic principle. As the mental parser makes its waythrough a sentence, it uses syntactic rules to analyze the sentence, and wheneverit uses a syntactic rule, it also uses the corresponding semantic principle to buildup an interpretation of the sentence. The literal meaning of "You are pulling myleg" can be constructed compositionally, and so different arguments for the verbare intersubstitutable, i.e., "You are pulling the part of my leg that hurts."Different superficial re-arrangements of the constituents are also acceptable, i.e.,"My leg is being pulled by you." The idiomatic sense, however, is not fullycompositional. "My leg" does not refer to my leg, but to me, and, unlike thecompositional case, listeners cannot make this interpretation of the noun phrasein isolation from the rest of the sentence. They need the expression as a whole torecognize that it has an idiomatic interpretation. Idioms are accordingly rela-tively fixed in syntactic and lexical form. Speakers who say, " You are pullingmy legs," or "You are pushing my legs," or "You're pulling my foot," have notquite mastered the idiom. Yet a fluent speaker might invent such expressions toconvey idiomatically some subtle aspect of the situation: "You are pulling mylegs" because you have told me two improbable stories. If a compositionalinterpretation is nonsensical in the context of the utterance, then the listener issupposed to check whether an idiomatic sense is listed and whether it makesbetter sense. In short, idioms are exceptions to the general rule of composi-tionality, and where the general rule fails, the list of exceptions (e.g., frozenidioms) is examined to see whether it provides a more appropriate interpretation.This sort of proposal has been advanced from many quarters, particularly byphilosophers, logicians, and those housebound cartographers who rely solely onintuition to draw maps of the mind.

The facts tell a different story. Why. for example, do idioms exist? Theirorigin is singularly mysterious in the logical account of language. It is not at allobvious why speakers should use expressions that cannot be understood in the

FOREWORD iX

normal way. Idioms should be rare exceptions comparable to slips of the tongue.Yet it is difficult to speak spontaneously without lapsing into idiomatic usage.Try it with a foreigner, and you will see that you are often aware of using anidiom only after the event. Even in explaining an idiom, you can find to yourexpense that you are using another. Hence, the first reason for the importance ofidioms is that they are pervasive.

One reaction to this pervasiveness is to argue that all usage is idiomatic.(Connectionists have a natural inclination to make such claims.) And this hypoth-esis has the advantage that it removes at a stroke the mystery of the origin ofidioms: They arise in the natural use of natural language. Yet fish are unaware ofthe water-that is to say, if all is idiom, why do we ever draw the distinctionbetween the literal and the idiomatic? Clearly, utterances with a literal interpreta-tion are just as pervasive as idioms. There may be a continuum from clear casesof literal usage to clear cases of idiomatic usage, but the two ends of thecontinuum are plain enough. The creative use of language-at any level fromphonology to pragmatics-is a natural part of discourse. Speakers-some morethan others-invent words and phrases to force us to pay attention, to amuse us,to astonish us, and to challenge us. And they create new ways to convey oldmeanings for the sheer joy of invention. But the creation of idioms also reflectsnew conceptions of the world, new ways in which individuals construct mentalmodels of the world, and new ways in which to convey their contents vividly. Itis through idioms-and I include those special cases that rhetoricians dignify astropes, such as metaphor, irony, metonymy, and synecdoche-that the trulycreative nature of human expression reveals itself. Idioms are the poetry of dailydiscourse. That is the second reason for their importance.

Alas, we are not all capable of idiomatic invention, but most of us do pick up,borrow, or steal the idioms of others. Our linguistic usage is full of second-handidioms, dead metaphors, and stale similes. We use these cliches without think-ing; and we understand them equally automatically. However, if we are supposedto try to make a literal interpretation before we seek an idiomatic one, then weshould understand literal meanings faster than idioms. In fact, as a number ofexperiments have shown, we can understand idioms just as fast as literal usages.For twenty years or more, psycholinguists have pursued an intensive examinationof how the mental parser works. Compositional interpretation is, in principle,straightforward. But how are idioms interpreted? Speakers certainly acquire aknowledge of familiar idioms, and this knowledge somehow enables them tocope as readily with idiomatic usage as with literal meaning. Idioms are easy, andthis surprising fact is the third reason for their importance.

We have the ability to speak in riddles. These riddles are neither constructednor interpreted in the normal way. Yet we use them so readily that we are usuallyunaware of their special character-unless we have the misfortune not to be anative speaker. We call these special riddles idioms. They are one of the manyways in which natural languages differ from those artificial languages that ex-

X FOREWORD

perts have designed on logical principles-mathematical calculi, computer pro-gramming languages, and systems of formal logic. Idioms are mysterious. Theyare pervasive, poetic, and easy. How they are understood is a mystery that thecontents of this book go some way to dispelling.

P. N. Johnson-Laird

Preface

According to the Oxford Dictionary (5th ed.), an idiom can be the "language of apeople or country," or a "form of expression peculiar to a language." It is in thesecond of the two senses, obviously related to each other, that idioms', alongwith metaphors, proverbs, indirect speech acts, etc., are part of the vast family offigurative expressions. The interest in these expressions has a longstanding tradi-ton. Within European structuralism, for instance, linguists have studied figur-ative expressions primarily in the attempt to shed light on the synchronic reg-ularities and the diachronic changes of languages (for a recent approach tolanguage changes, cf. Sweetser, 1991). In a similar vein, anthropologists havedevoted much of their attention to studying how a figurative expression is formedor why it becomes an appropriate description of an event in one speech commu-nity, but not in another (Basso, 1976; Fernandez, 1991; Holland & Quinn, 1987;Ruwet, 1983). These lines of research, aiming at elucidating the linguistic andsocial sources of figurative language, have successfully shown that any facts,beliefs, cultural models, natural events, etc. shared by the members of a speechcommunity at a given time can give rise to figurative expressions (Ammer, 1989;Makkai, 1987). The introduction of railways, electricity, and other technologicalinnovations, for instance, has enriched the figurative basis of many languagesproviding a whole new set of ways of describing reality, which was not available,say, in seventeenth century France, where many figurative expressions wererelated to religion (Ullmann, 1962).

More recently, the interest in the origins of figurative language has receivedfurther impulse thanks to a new approach that sees the substratum of figurative

We will use "idiom" and "idiomatic expression" as synonyms.

xi

Xii PREFACE

expressions in people's cognitive structure (e.g., Holland & Quinn, 1987; Nayak& Gibbs, 1990; Sweetser, 1991). In this perspective, the historical origins offigurative expressions, whether or not they are still transparent to the speakers ofa community, provide useful information on how people organize their concep-tual and lexical knowledge, and establish connections across domains. A well-known example of this sort is the relation between the temporal and spatialdomain, where many concepts of the former are borrowed from the latter, asillustrated by the numerous spatial terms used-at least in English-for referringto time.

In this long and uninterrupted tradition, figurative language has typically beenviewed as a relatively homogeneous topic. No doubt, there are differencesamong, for instance, a metaphor and a proverb, but both are likely to havesimilar origins and serve similar purposes within a linguistic community. More-over, a widely shared assumption is that, theoretically, the most prominent formis metaphor, whereas the other expressions are mostly derivative topics. Idioms,in particular, have often been considered "dead metaphors"-i.e.., expressionsthat were once innovative, but are now conventionalized and frozen-and hencescarcely relevant in comparison with metaphor.

Only recently has this belief come to be challenged, and idiomatic expressionscome to be considered for their own sake. This book takes up this perspective,and although several chapters discuss the relation between idioms and otherforms of figurative language-notable, metaphor-its main focus is on aspectsthat are peculiar to idioms. Thus, the book deals primarily with questions such ashow idioms are mentally represented, understood, and acquired by children,what are the neurological structures involved in their comprehension, how cantheir syntactic behavior be explained, and what is their place in our languages.

This approach is grounded in the conviction that problems of representation,parsing and interpretation are different for metaphors and idioms, each of whichhas specific properties and characteristics, and requires, therefore, different ex-planations. This is not to deny the unquestionable relation between the twoforms. To the contrary, one fascinating aspect of idioms-extensively exploredin the book-is precisely the fact that idioms appear to be at a crossroad, sharingaspects of different linguistic objects. Thus, although lacking the semanticrichness and flexibility of metaphors, idioms share some of the features of that"most luminous" and vital trope. For instance, individual words in idiomaticexpressions, as in metaphors, do not generally have the same meaning they havein literal strings. As Levinson (1983) has noted for metaphors, in order to makesense of both types of expressions one must take into account the "connotativepenumbra" of their words. However, unlike metaphors, whose interpretation isbuilt compositionally, the meaning of many idiomatic expressions seems to be-come available, as the meaning of lexical items, through processes of retrievalfrom memory. Yet again, unlike actual words, idioms have a syntactic structurethat at times is frozen, but on occasions is very flexible and can be modified in

PREFACE Xiii

various ways, depending also upon the extent to which the figurative meaning ofthe string is related to its literal meaning (Cutler, 1982; Gibbs & Nayak, 1989;Gibbs, Nayak, Bolton, & Keppel, 1988; Nunberg, 1978; Wasow, Sag, & Nun-berg, 1983).

Indeed, as pointed out by almost all of the authors in this book, idioms arevery elusive, and the difficulty of exactly characterizing them is perhaps one ofthe reasons why relatively little attention has traditionally been accorded to theseexpressions, in spite of their unquestionable relevance, which resides in at leasttwo considerations: Idioms are a clear challenge to current compositional modelsof language comprehension, and their use in the language is so widespread tojustify Searle's (1975) informal rule of conversation: "Speak idiomatically un-less there is some good reason not to do so."

The multifaceted nature of idioms has, among its consequences, the fact that itis virtually impossible for any single approach or methodology to fully capture it.Moreover, our current knowledge of these expressions is such that the picturethat emerges from their study is still rather fragmentary. Both these aspects arereflected in the book, and although linguistic, computational, psychological, andneuropsychological approaches are all present, providing an interdisciplinary,cognitive science perspective on the study of idioms, no attempt has been madeto force the different trends of research into a coherent framework. Instead, thechapters offers an updated account of many of the problems that are currentlydiscussed, along with the explanatory attempts produced so far.

The book is divided into three parts. Part I is dedicated to the interpretation ofidioms and to the relations between their literal and the figurative meanings.Glucksberg's claim is that idioms are not unanalyzable wholes, but can bedecomposed into linguistic and conceptual elements, even though the literalmeanings of the linguistic elements alone are insufficient, and other sources ofmeaning-in particular, stipulated and allusional-must also be considered inorder for the complete interpretation of these expressions to be obtained.

The notions of literal and metaphorical meanings are also discussed in thechapter by Cacciari. She argues that for many idioms the meanings of theirindividual words and the rhetorical structure of the expressions-whether meta-phorical, analogical, or other-are both available to people. This information isnot always used, but is employed when required either by a specific task, or bythe need to interpret unfamiliar idiomatic expression.

Gibbs' chapter is aimed at demonstratiang the fallacy of the "dead metaphorview of idiomaticity." The author argues that many idioms are "very much alivemetaphorically", and that people are able to make sense of the figurative mean-ing of many unfamiliar idioms precisely because they can rely on the meta-phorical interpretation of these expressions. Finally, Flores d'Arcais examinesthe processes of comprehension of an idiomatic phrase with respect to its famil-iarity, its uniqueness point-i.e., the point at which the idiom becomes uniquelyidentifiable-and its syntactic flexibility. He also presents data suggesting that

XiV PREFACE

people interpret unfamiliar idioms much like metaphors, relying whenever pos-sible on similarities with more familiar expressions.

Part 11 of the book deals with how children learn to use idiomatic expressionsand how these are represented and processed by adults. In her chapter on acquisi-tion, Levorato argues that learning to use an idiom is not the same as learning toassociate the string with its meaning, as it would be if idioms were acquired likewords. Rather, it is a process that requires the development of figurative compe-tence, and is achieved in different stages during which the ability to comprehendand produce idioms grows in parallel with the child's increasing mastery oflinguistic and communicative abilities.

Botelho and Cutler's chapter presents data on memory for Portuguese idiomswith and without a literal counterpart. The authors argue that their results supportthe lexical representation hypothesis originallyy proposed by Cutler and Swinneyin 1979), one of the most influential psycholinguistic models of idiom processingand representation. In contrast to this view, Tabossi and Zardon claim that idiomsare mentally represented not as lexical items, but as configurations of words,much like poems. Accordingly, the meaning of these expressions is retrieved notas the meaning of individual words, but triggered only after sufficient informa-tion is available to the listeners to recognize the idiomatic "configuration."

Colombo's paper deals with an old issue in psycholinguistic research-name-ly, the ambiguity of idioms and the effects that context has on the biasing of theirliteral and figurative interpretation. The results of her study appear to indicatethat the idiomatic meaning of an ambiguous string is activated in a context thatbiases that meaning. No such activation is found, however, in a neutral context orin a context that biases the literal meaning of the string.

In the last chapter of the section, Peterson and Burgess argue for the relevanceof neurolinguistic as well as psycholinguistic evidence in the study of idioms. Infact, both types of data presented in their chapter appear to converge, indicating adissociation between the syntactic and semantic processing of idiomatic ex-pressions. In particular, the syntactic analysis of an idiom is computed evenwhen such an analysis is no longer used to contruct the literal interpretation of thestring.

In Part 111, the chapter by Stock, Slack, and Ortony and that by Pulman bothtackle one of the best known and least understood problems involving idioms:their flexibility. As Pulman correctly points out, the major difficulty with idiomsis not "that their meaning cannot by derived compositionality, . . . because thiscould by overcome by treating them as multi-word lexical items . . . The prob-lem is that unlike (most) lexical items, most) idioms have considerable internalstructure which seems to interact with the usual productive syntactic and seman-tic mechanisms of a language."

Stock et al. address this thorny problem in a computational model where theflexibility of an idiom, or the lack thereof, is determined by whether or not thethematic structure of its linguistic form (e.g., "kick the bucket") is equivalent tothe thematic structure of its associated semantic representation (e.g., "die").

PREFACE XV

In a similar vein, Pulman argues that the flexibililty of idioms is not andshould not be explained by a theory of syntax, but depends on the semantics ofthe idioms and on the contextual interpretation of the utterances in which theyoccur. Given a sentence, the processing system parses it and, applying composi-tional semantic rules, gives the logical forms of the sentence which in turn canserve as input to an inferential mechanism that has access to meaning postulatesand idiom rules. These can add figurative interpretations to the string. The sets ofliteral and figurative interpretations are then evaluated contextually.

Fellbaum's chapter is more specific in scope. After an analysis of the distribu-tion and role of determiners in idiomatic strings such as "spill the beans" or"have an axe to grind," Fellbaum discusses the results of the investigation in thelight of a theory of compositionality.

In the final chapter, Makkai explores universal constraints on the formation ofidioms, and discusses the role that this process fullfils in a language. Makkaiclaims that language develops through phases, the last of which is characterizedby the arbitrary attribution of sound strings to concepts. However, when conceptsdevelop even further, a "point of saturation" is reached in the process of desig-nating new concepts and objects arbitrarily. The speech community starts then toborrow images and similes for what is to be expressed, and recombines existingsigns in new ways, in this way creating idioms.

This brief presentation should suffice to give the reader a flavor of the varietyof topics that are currently under debate, and of the different approaches that co-exist in the field. In fact, as already pointed out, many aspects concerning idiomsremain controversial, and perhaps many questions are still waiting to be asked inthe appropriate way. However, along with the differences, the chapters in thisbook also suggest that, at least on some issues, convergent attitudes are startingto emerge. For instance, the traditional, noncompositional approach to idiomaticmeaning, once unchallenged, has recently been criticized on several grounds,and the view that the meanings of the constituent words of an idiom do play arole in its comprehension is now winning a growing consensus among cognitivescientists. We hope that this book will contribute to the further development ofinterdisciplinary research, offering a useful tool to all those interested in thestudy of idioms and in the better comprehension of language, whether figurativeor not.

Cristina CacciariPatrizia Tabossi

REFERENCES

Ammer, C. (1989). It's raining cats and dogs . . . and other beastly expressions. New York: Laurel

Books.Basso, K. H. (1976). "Wise Words" of the Western Apache: Metaphor and semantic theory. In K.

Xvi PREFACE

H. Basso (Ed.), Meaning in anthropology (pp. 93-121). Albuquerque: University of New Mex-ico Press.

Cutler, A. (1982). Idioms: the older the colder. Linguistic Inquiry, 13, 2, 317-320.Fernandez, J. W. (1991). (Ed.). Beyond metaphor. The theory of tropes in anthropology. Stanford,

CA: Stanford University Press.Gibbs, W. R., Nayak, N. P., Bolton, J. L., & Keppel, M. (1988). Speakers assumptions about the

lexical flexibility of idioms. Memory & Cognition, 17, 1, 58-68.Gibbs, W. R., & Nayak, N. P. (1989). Psycholinguistic studies on the syntactic behavior of idioms.

Cognitive Psychology, 21, 100-138.Holland, D., & Quinn, N. (1987). (Eds.). Cultural models in language and thought. New York:

Cambridge University Press.Levinson, S. C. (1983). Pragmatics. London: Cambridge University Press.Makkai, A. (1987). Idiomaticity and phraseology i.n post-chomskian linguistics. The coming-of-age

of semantics beyond the sentence. Semiotic&, 64, 1-2, 171-187.Nayak, P. N., & Gibbs, W. R. (1990). Conceptual knowledge in the interpretation of idioms. Journal

of Experimental Psychology: General, 119, 115-130.Nunberg, G. (1978). The pragmatics of reference. Bloomington, Indiana: Indiana Linguistic Club.Ruwet, N. (1983). Du bon usage des expressions idiomatiques. Recherches Linguistiques, 11, 5-84.Searle, J. (1975). Indirect speech acts. In P. Cole & J. L. Morgan (Eds.), Syntax and semantics

speech acts (pp. 59-82). New York: Academic Press.Sweetser, E. (1991). From etymology to pragmatics. Metaphorical and cultural aspects of semantic

structure. Cambridge: Cambridge University Press.Swinney, D. A., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journal of

Verbal Learning and Verbal Behavior, 18, 523-534.The Concise Oxford Dictionary, Fifth Edition. Oxford: The Clerendon Press.Ullmann, S. (1962). Semantics: An introduction to the study of meaning. Oxford: Basil Blackwell &

Mott.Wasow, T., Sag, I., & Nunberg, G. (1983). Idioms: An interim report. In S. Hattori & K. Inoue

(Eds.), Proceedings of the XI1th International Congress of Linguistics (pp. 102-105). Tokyo.

List of Contributors

Teresa Botelho da Silva Department of Linguistics, University of Cambridge,Sidewick Avenue, Cambridge CB3 9DL United Kingdom.

Curt Burgess Department of Psychology, 1419 Life Science Building, Univer-sity of California, Riverside, CA 92521-0426.

Cristina Cacciari Dipartimento di Psicologia, Viale Berti-Pichat 5, 40127Bologna, Italy.

Lucia Colombo Dipartimento di Psicologia Generale, Piazza Capitaniato 3,35139 Padova, Italy.

Anne Cutler MRC Applied Psychology Unit, 15 Chaucer Rd., CambridgeCB2 2EF United Kingdom.

Christiane Fellbaum Cognitive Science Laboratory, Princeton University,221, Nassau Street, Princeton, NJ 08542.

Giovanni Battista Flores d'Arcais Dipartimento di Psicologia dello Sviluppoe della Socializzazione, Via Beato Pellegrino 26, 35137 Padova, Italy.

Ray W. Gibbs Program in Experimental Psychology, Clark Kerr Hall, Univer-sity of California, Santa Cruz, Santa Cruz, CA 95064.

Sam Glucksberg Department of Psychology, Green Hall, Princeton Univer-sity, Princeton NJ 08544- 1010.

XViii LIST OF CONTRIBUTORS

Maria Chiara Levorato Dipartimento di Psicologia dello Sviluppo e dellaSocializzazione, Via Beato Pellegrino 26, 35137 Padova, Italy.

Adam Makkai M/c 237, University of Illinois at Chicago, POB # 4348,Chicago IL 60680.

Andrew Ortony Institute for the Learning Sciences, 1980 Maple Avenue,Evanston, IL 60201.

Robert Peterson Department of Psychology, Indiana University, Bloom-ington, IN 47405.

Steve G. Pulman SRI International, Cambridge Computer Science Research,Millers Yard, Mill Lane, Cambridge C132 IRQ United Kingdom.

Jon Slack IRST, 38050 Povo (Trento), Italy.

Oliviero Stock IRST, 38050 Povo (Trento), Italy.

Patrizia Tabossi Facolta di Letteu e Filosofia, Via Savonarola 38, 44100Ferrara, Italy.

Francesco Zardon Dipartimento di Ps icologia, Viale Berti-Pichat 5, 40127Bologna, Italy.

IDIOM INTERPRETATIONAND THE LITERALFIGURATIVE DISTINCTION

1

Idiom Meanings andAllusional Content

Sam GlucksbergPrinceton University

People who live in glass houses should not throw stones.

For most readers, this proverb should elicit the experience of apprehendingseveral kinds of meanings simultaneously. The meaning of the sentence itself-the literal meaning-would be apprehended by anyone who reads English flu-ently. In addition, readers familiar with the proverb will apprehend immediatelya second kind of meaning, the meaning of the proverb. The expression is not justabout glass houses and stones, but also about the vulnerability of people criticiz-ing others for faults that they themselves have. The idiomatic meaning itselfderives from the allusion to the glass house as a metaphor for vulnerability.Despite being able to determine both the literal and idiomatic meanings of theproverb, however, few readers experience the most important meaning of all:What did 1, the writer, intend by my use of this proverb? In the absence of anyrelevant contextual information, the proverb can be understood only as an exam-ple or illustration of some point, which of course it is.

The proverb is one example of the class of expressions that mean somethingother than their constituent words and phrases. At one end of the continuum arephrases such as by and large, which seem to be nothing more than long words.Furthermore, not only does this expression mean something other than its constit-uents, its meaning seems to bear no relation to those constituents. At the otherextreme are familiar proverbs and idioms that allude to apocryphal events, suchas residents of glass houses throwing stones, people carrying coals to Newcastle,and farmers locking barn doors after horses have been stolen. In these cases, themeanings of the constituents are relevant, but the meaning of each expression isnot just something other than the meanings of the constituent parts. The ex-

3

4 GLUCKSBERG

pression's meaning is also something more than the meanings of the parts. Theexpression itself alludes to an archetypical case of the class of events that ittypifies. It does not matter one whit whether anyone actually threw any stoneswhile living in a glass house, or carried coal to the English city of Newcastle, orever locked a barn door after having horses stolen. These expressions are, inessence, metaphors for the general situations or events that they typify.

In between the wordlike by-and-large idioms and the metaphorlike coals-to-Newcastle idioms are those that can have quite variable relations between theirconstituent and idiom meanings. Like by and large, phrasal idioms such as kickthe bucket have meanings that bear no discernible relation to their idiomaticmeanings, in this case "to die." Like coals to Newcastle, other phrasal idiomssuch as spill the beans bear a somewhat metaphorical relation to their idiomaticmeanings, in this case divulge secrets. Given the diversity among the idiomspeople use in everyday discourse, it should not be surprising that theories ofidiom comprehension are diverse as well.

Idiomatic Meaning: Direct Access VersusCompositional

Two classes of models have been proposed for idiom comprehension. Reflectingthe characteristics of such idioms as by and large and kick the bucket, one classtreats idioms as expressions that have meanings that are stipulated arbitrarily.According to this class of models, idioms are understood simply by retrieving themeaning of an idiom as a whole. I refer to this type of model as a direct look-upmodel. The second class of models reflects the characteristics of such idioms ascarrying coals to Newcastle. The meanings of these kinds of idioms are notarbitrary. The relation of coals to Newcastle is a matter of historical fact, and theliteral act of carrying coals to the coal-mining center of Newcastle is a ster-eotypical instance of uselessly bringing something to some place. According tothis second class of models, idioms are understood by ordinary linguistic pro-cessing combined with a pragmatic interpretation of the use of the expression indiscourse contexts. I refer to this class of models as compositional.

Direct Look-Up. Three versions of direct look-up models have been pro-posed: (a) the idiom list hypothesis (Bobrow & Bell, 1973), (b) the lexicalizationhypothesis (Swinney & Cutler, 1979), and (c) the direct access hypothesis(Gibbs, 1984). All share the assumption that idiom meanings are apprehended bydirect memory retrieval, not by linguistic processing. The three differ in rela-tively unimportant ways. Bobrow and Bell proposed that idioms are representedin a mental idiom list, that is, an idiom lexicon that parallels the mental wordlexicon. Idiomatic meanings are sought when a linguistic analysis fails to yieldan interpretable result. When linguistic analysis fails, people turn to a search ofthe idiom list, and if the linguistically recalcitrant expression is found, then the

1. IDIOM MEANINGS AND ALLUSIONAL CONTENT

idiom meaning is taken as the intended meaning. This model is rejected easily bythe robust finding that idioms are understood at least as quickly as comparableliteral expressions (Gibbs, 1980; Ortony, Schallert, Reynolds, and Antos, 1978).If expressions must always be analyzed literally before any idiomatic meaningsare sought, then literal meanings should always be understood more quickly thanidiomatic ones. Contrary to Bobrow and Bell, the literal meanings of conven-tional idiomatic expressions are never understood more quickly than their idi-omatic ones.

Swinney and Cutler's (1979) lexicalization hypothesis accounts quite nicelyfor the relative ease of understanding familiar idioms. Idioms are representedsimply as long words, together with all the ordinary words in the mental lexicon.When a familiar idiomatic expression is encountered, linguistic processing pro-ceeds normally. Lexical access, of course, proceeds as part of linguistic process-ing, and lexicalized phrases such as by and large or kick the bucket are routinelyfound in the mental lexicon along with their constituent words, by, and, large,and so on. Which of the two meanings-literal or idiomatic-is apprehendedfirst depends on the relative speed with which full linguistic processing andlexical/idiom access can be completed. Normally, idiom access will be com-pleted more quickly because it does not require the lexical, syntactic, and seman-tic processing required for full linguistic analysis. Thus, familiar idioms will beunderstood more quickly than comparable literal expressions.

Gibbs' (1984) direct access proposal is an extreme version of the Swinney andCutler (1979) model. Rather than posit a race between idiom-meaning access andlinguistic processing, Gibbs argued that linguistic processing may be bypassedentirely if an expression is recognized immediately as an idiom. In essence,idiom-meaning access may be so rapid as to obviate any linguistic analysis at all.Gibbs himself soon adopted a more compositional view of idiom comprehension,based in part on some of his own observations of lexical and syntactic phe-nomena involved in idiom use.

Compositional. As we have already seen, idioms can vary from apparentlyunitary phrases (e.g., by and large) to expressions whose idiomatic meaningsderive jointly from their literal meanings and allusional content (e.g., carry coalsto Newcastle). Nunberg (1978) tried to capture this variability by proposing thatidioms can be ordered along a continuum of compositionality. Ordinary languageis, by definition, compositional. The meaning of any given linguistic expressionis determinable from the meanings of its constituent parts and the syntactic andsemantic relations among those parts. The meaning of any idiom, in the standardview, is determinable entirely from its stipulated meaning, whether that meaningis represented in a special idiom list or simply as part of the mental lexicon.

As usual, a simple dichotomy fails to capture natural complexity. In an exten-sion of Nunberg's (1978) original proposal, Gibbs and his colleagues have shownthat people can reliably judge degrees of compositionality of idioms (Gibbs,

5

6 GLUCKSBERG

Nayak, Bolton, & Keppel, 1989; Gibbs, Nayak, & Cutting, 1989). Idioms suchas spill the beans, for example, are considered relatively compositional (inGibbs' terms, analyzable), in contrast to idioms such as kick the bucket, whichare considered noncompositional. In either case, linguistic processing proceedsin parallel with direct idiom-meaning look-up, with direct look-up usually beingfaster than full linguistic analysis.

A somewhat similar proposal was offered by Cacciari and Tabossi (1988).Linguistic processing and idiom look-up can occur in parallel, but idiom look-upcannot begin until the idiom itself is recognized as a configuration, that is, as aunitary expression with a meaning beyond that of its constituents. This modelalong with the race models mentioned earlier seem to fit most closely with whatwe know about idiom processing. In the next section, important idiom phe-nomena are considered in the context of the issue of look-up versus composi-tionality.

HOW IDIOMS ARE UNDERSTOOD

Idioms as Long Words

The primary evidence for direct look-up of idiom meaning is the relative speed ofidiom comprehension. Idioms are understood more quickly in their idiomaticsenses than in their literal senses. The to-die meaning of kick the bucket, forexample, is understood more quickly than the literal meaning of striking a pailwith one's foot (Gibbs, 1980). Similarly, it takes less time to understand theexpression spill the beans than to understand the literal paraphrase, tell thesecrets (McGlone, Glucksberg, & Cacciari, in press; see also, Ortony et al.,1978). These data suggest that an idiom's meaning may be retrieved from memo-ry without full linguistic processing, on the assumption that direct memoryretrieval takes less time than would standard linguistic processing.

Idioms as Linguistic Expressions

Whatever else they may be, idioms are composed of words that in turn formphrases and sentences. In general, people cannot inhibit their language-process-ing system. If someone attends to a word, for example, then they cannot ignorethat word's meaning. Even if people are asked explicitly to ignore a word'smeaning, as in Stroop's classic color-naming experiment, the meaning stillcomes through. People who try to name the color of the ink that a color name isprinted in are delayed momentarily when the color name and ink color differ(e.g., when the word red is printed in green ink; Stroop, 1935). Given theautomaticity of the language-processing system, it should not be surprising tofind evidence for the ubiquity of lexical and syntactic operations during idiomcomprehension.

1. IDIOM MEANINGS AND ALLUSIONAL CONTENT 7

Phrases Versus Words. Perhaps the most compelling case for idioms as longwords can be made with such idioms as by and large. Like a long word, suchidioms can be negated, as in:

Tom: By and large, the economy seems to be doing well.Ned: Not so by and large: Have you seen the latest unemployment

figures?

Unlike a word, however, such idioms can be negated internally, as in:

Ned: By but not so large! Have you considered....

If the string by and large were indeed nothing more than a long word, thensubstituting but for and, together with inserting two additional words, not and so,should produce an unacceptable string. Not only is the string acceptable, it isperfectly interpretable. Phrases such as by and large, however much they mightbehave like long words, are still phrases and treated as such.

Do Words Matter? Semantic Compatibility Effects. Some phrasal idiomsseem odd when synonyms are substituted for the original words, and indeed maynot even be recognized as idioms; for example, people rarely realize that boot thepail is a paraphrase of kick the bucket (Gibbs, Nayak, & Cutting, 1989). Otherphrasal idioms, especially those that are judged to be compositional, can survivelexical substitutions, but the substitutions are constrained jointly by the idiom'smeaning and the semantics of the words themselves. The idiom break the ice, forexample, refers to a more or less discrete event that results in a relaxation of astiff, awkward, chilly social situation. Substituting the word crack for break inthis idiom is relatively acceptable. In contrast, the words crush, grind, or shavewould not be acceptable in this idiom, even though these actions are perfectlyappropriate to the actual object, ice. Can the proverbial ice be melted? Perhaps,but only if a gradual change in the social atmosphere were involved. Theseexamples illustrate the potential role of literal word meanings in idiomatic useand comprehension. Lexical substitutions are not only possible but also aresemantically constrained precisely because idioms must be processed lin-guistically, even when such processing is not necessary for determining theidiom's meaning.

Semantic Constraints on Idiom Use. Idioms such as kick the bucket tend toresist lexical substitutions. Nevertheless, the semantic properties of their constit-uents may still play important roles in use and in comprehension. Even thoughthere is no apparent relation between the meanings of the words kick and bucketand the concept "to die," word meanings and idiomatic meaning may stillinteract to guide and constrain this idiom's use. On the one hand, our understand-ing of what it means to die guides and constrains how the idiom kick the bucket

8 GLUCKSBERG

may be used. People can die silently, and so it makes sense to say, "He silentlykicked the bucket." People cannot die "sharply," so even though one can kicksharply, one cannot say "He sharply kicked the bucket." On the other hand, themeanings of the words kick and bucket can also play important roles. Kicking isa discrete act, and so, even though one can say, "He lay dying all week," onecannot say, "He lay kicking the bucket all week" (Wasow, Sag, & Nunberg,1983). This is because the only way one can kick a bucket all week is to kick itover and over again, but one cannot die over and over again.

How real is the bucket? Once used in discourse, the proverbial bucket behavesjust as would any other discourse referent, as shown by its availability as ananaphoric referent. Consider the following conversational fragment:

George: Did the old man kick the bucket last night?Edward: Nah, he barely nudged it.

In this context, the relation of nudge to kick and the use of the pronoun it to referto bucket are clearly interpretable. Barely nudging as compared to kicking thebucket denotes not even coming close to kicking it, and so this variant is taken tomean not even close to dying. In this example, the concept of death and thesemantics of the idiom's constituents jointly constrain idiom use and comprehen-sion, reflecting again the joint operation of idiom meaning and the languageitself.

Idiom Variants: Semantic Productivity. When familiar idioms are used asvariants of their canonical forms, the discrepancy between original and newwordings may form the basis of newly created idiom meanings. If, for example,someone were to say crack the ice instead of break the ice to refer to a change ina social situation, no particular communicative intent would be inferred. Crackand break would be seen as mere stylistic variants of one another. If, on the otherhand, someone were to say shatter the ce, then this would not be interpreted as amere stylistic variant. Instead, the difference between the meaning of break andthe meaning of shatter creates a new idiomatic meaning, something like "breakdown an uncomfortable and stiff social situation flamboyantly in one fellswoop." Thus, this is not simply an example of lexical flexibility, it is anexample of semantic productivity.

Such semantically productive idiom variants appear in everyday conversationand in the media. One striking example appeared in a New York Times article onthe rise and fall of the Wall Street firm Drexel Burnham Lambert (Drexel).Drexel had made a fortune on junk bonds and then found themselves seriouslyshort of cash. Before declaring bankruptcy, the firm's assets were distributedamong the senior executives in the form of very substantial cash bonuses. As adirect result of this bonus distribution, Drexel's cash reserve was depleted,forcing the firm into bankruptcy. In this context, the meaning of this twist on a


familiar idiom is clear: "Drexel's senior executives, not content with collectingone golden egg after another, seem to have insisted then on eating the goose."Similarly, Donald Barthelme's title for an essay on contemporary literature-"Convicted Minimalist Spills Bean"-makes perfect sense to those who knowof his reputation as a minimalist writer. Even young children spontaneously useidioms productively in discourse, once the original idioms have been learned.One 5-year-old girl learned the expression spill the beans as meaning "tell asecret." Later that same day, she told her father, "Don't throw the beans toRebecca! She's not supposed to know!" (Greenberg-Concool, 1990).

Direct Access and Linguistic Processing Reconsidered

The evidence for direct access is straightforward yet inconclusive: (a) Familiaridioms in their original, canonical form are understood more quickly than com-parable literal expressions; and (b) Familiar idioms in canonical form also areunderstood more quickly than their variants (McGlone, Glucksberg, & Cacciari,in press). The expression didn't spill the beans, for example, is understood morequickly than the relatively novel expression didn't spill a single bean. Takentogether, these data suggest but do not force the conclusion that familiar idiomsneed not be processed linguistically because their meanings can be retrieveddirectly from memory. In contrast, nonidiomatic expressions, as well as variantsof familiar idioms, presumably must be processed linguistically because theirmeanings are not stipulated and stored in memory.

At the same time, familiar idioms must be processed linguistically eventhough initial meaning access may be quite independent of linguistic analysis.Evidence that even familiar idioms are fully analyzed includes the reciprocalconstraints of idiom meaning and literal meaning on idiom use, and the ability ofpeople to understand idiom variants quite readily. The question remains, how-ever, how do people determine the meaning of idiom variants? An adequateanswer to this question might provide the answer to the question of how peopleunderstand idioms more generally.

UNDERSTANDING IDIOMS AND THEIRVARIANTS: A PARADIGM CASE

How do people manage to understand an expression such as "He didn't spill asingle bean"? Clearly, the meaning of this expression cannot be stored in memo-ry because the expression is novel, albeit based on the familiar idiom spill thebeans. If a variant idiom's meaning is not available to be retrieved from memory,then the meanings of the constituent words must be used in some fashion todetermine the variant's meaning. There are at least two ways that word meaningscould be used. One way would involve comparing the meanings of the original

10 GLUCKSBERG

and variant idiom constituents, determining the relation between those meanings,and inferring, by analogy, the meaning of the variant with respect to the meaningof the original. Such a strategy would involve at least the following six discreteoperations:

1. Recognize novel idiom as a variant of a conventional idiom.2. Retrieve meaning of original idiom.3. Identify word meanings of both variant and original idioms.

4. Compare the word meanings of the two idiom forms.5. Identify the relation(s) between those word meanings.6. Take this relation(s) between the word meanings to infer, by analogy, the

relation(s) between the meanings of the original and variant idioms.

If, for example, the substituted words were antonyms, then the variant idiom'smeaning would be taken to be the opposite of the original, as in got up on theright side of the bed versus got up on the wrong side of the bed. If the relationbetween the substituted words is one of quantity, then this would be taken as therelation between the variant idiom and the original, as in spill a single beanversus spill the beans. More complex relations are also possible, as in the DrexelBurnham example, where the proverbial golden goose can be said to be eateninstead of killed to imply unseemly greed and gorging oneself in addition tosimply destroying a source of wealth. Similarly, to say that "we'll jump off thatbridge when we come to it" instead of "we'll cross that bridge when we come toit" suggests self-destructive or suicidal behavior in addition to lack of foresightor planning.

This kind of sequential, multistep model is perfectly compatible with thetraditional view of idioms as long words (or as phrases) whose meanings aresimply retrieved from a mental idiom lexicon. Individual word meanings, ofcourse, must be used when conventional idioms are varied in some way, butword meanings still would play no role in understanding familiar idioms in theircanonical form. One implication of this view is that variant idioms should takeconsiderably longer to understand than idioms in their original forms. Originalidioms require only one operation, retrieval of a meaning from a stored list ofmeanings (Operation 2 from previous six-item list). Variant idioms, on thismodel, would require at least the five additional operations listed earlier. Even ifsome of these operations could be done in parallel (e.g., retrieving the meaningof the original idiom and activating word meanings), variant idioms still shouldtake more time to process than original idioms, and certainly more time toprocess than literal expressions of comparable length and complexity.

A much simpler alternative model uses the meanings of the words of thevariant idiom to arrive at the variant's meaning. The core assumption of thismodel is that the words of familiar idioms have become polysemous through


frequent use in idiom contexts. The verb and noun in the idiom spill the beans,for example, have at least two meanings: their default context-free literal mean-ings, and the meanings that are induced by the idiom context. In nonidiomaticcontexts, the verb spill will have the meaning "to be lost from a container" andthe word beans the meaning "edible legumes." In the idiom context, these wordshave a dual meaning, retaining their literal meanings but also acquiring theidiomatic meanings of "reveal" and "information-that-should-have-been-kept-confidential." With repeated usage, such idioms become able to induce poly-semy, adding the idiomatic meanings to each word's set of possible meanings.

Once this property has developed for any particular idiom and its constituentwords, variants of the idiom could be processed just as one would process anyother phrase or sentence: by accessing the contextually appropriate word mean-ings and performing ordinary linguistic analyses on the words and their relationswith one another. In the case of familiar idioms, this will result in at least twoproducts: the literal meaning and the idiomatic meaning (including the idiomaticsenses of the words of the idiom). Because this model depends on the assumptionof phrase-induced dual meanings, we refer to it as the phrase-induced polysemy(PIP) model of idiom comprehension. I

The PIP model has at least two interesting advantages over the standard idiomlook-up model. First, it allows for rapid and easy comprehension and productionof variant idioms. Second, it is parsimonious in that the same model wouldaccount for both conventional and variant idiom processing. The PIP model canbe tested by assessing the relative comprehension time for variant idioms andcomparable literal expressions. According to the PIP model, variants such asdidn't spill a single bean should take more time than original idioms such as spillthe beans precisely because the variants must be analyzed as if they were literalstrings. If this is so, then a variant idiom such as didn't spill a single bean shouldtake no longer to process than its literal paraphrase didn't say a single word. Incontrast, the standard model assumes that variant idioms require several moreprocessing operations than comparable literal expressions. Therefore, the stan-dard model predicts that variant idioms should take longer to process than theirliteral paraphrases.

Following this logic, McGlone et al. (in press) compared comprehensiontimes for original and variant idioms and their literal paraphrases. Original idi-oms were faster than both their literal paraphrases and their variants. Moreinteresting, variant idioms were understood just as quickly as their literal para-phrases: For example, didn't spill a single bean was understood as quickly asdidn't say a single word. How can this pattern of results be explained? Therelative speed of understanding original idioms can be accounted for by assumingthat once an idiom is recognized, then its meaning can be accessed directly from

'As Wasow et al. (1983) pointed out, "a full account of idioms must await a theory of meaningtransfers" (p. 114). The PIP model is a step toward explicating such a theory.

1 1

12 GLUCKSBERG

memory, often before linguistic processing is completed. This is essentially arestatement of the lexical representation and direct access hypotheses (Gibbs,1980; Swinney & Cutler, 1979). Additional assumptions are needed, however, toaccount for the finding that variant idioms can be understood as rapidly as theirliteral paraphrases. The most parsimonious processing assumption that accountsfor the entire pattern of results is that variant idioms, like literal expressions,require linguistic processing, whereas original idioms do not (at least for initialinterpretation).

STRING MEANINGS AND CONSTITUENTWORD MEANINGS

Consider how idiom meanings are acquired in the first place. As the standardview suggests, idiom meanings are acquired by stipulation. The meaning of anidiom is learned simply as an arbitrary relation between a phrase and its meaning,just as the meaning of a word is learned as an arbitrary relation between alinguistic unit and its referent. The meaning of spill the beans, for example, ismemorized, roughly, as "reveal the secrets." This paraphrase is not precise, noris it limited to the particular words reveal or secrets. Other literal paraphraseswould suit as well, such as "divulge/tell/inform about" [the] "informa-tion/plans/confidential stuff, and so forth. Although the wording may vary, thebasic conceptual referents and the relations among them are, of course, invariant.With repeated use, it would not be surprising if the pairing of a word such as spilland the concept of "divulge" would come to be represented in lexical memory.In the context of a phrase containing the word beans, the idiomatic meaning ofspill would be activated automatically, just as the context-appropriate meaning ofany polysemous word is activated automatically (Tabossi, 1988). More specifi-cally, the two words, spill and beans, come to function as minimal contexts forone another, making available the dual meanings of spill and beans duringcomprehension.

After such idiomatic word meanings have been acquired, then the meanings offamiliar idioms could be either retrieved or generated. Retrieval would be accom-plished by direct access of the stipulated idiomatic meaning. Generation wouldbe accomplished by ordinary linguistic processing, using the contextually appro-priate idiomatic meanings of the idiom's constituent words.2 Because directaccess can be accomplished more quickly than linguistic processing, familiaridioms can be understood more rapidly than comparable literal expressions,because such idioms do not require linguistic processing whereas ordinary literal

2The polysemy of idiom constituents, in fact, aIs recognized by lexicographers. For example, theRandom House Dictionary of the American Language lists, as one entry for the word spill, "todivulge, disclose or tell" (p. 1266).


expressions do. However, if direct access fails, as when memory retrieval mightfail, then an idiom might still be understood via linguistic processing. Becausevariant idioms use words that do not match the canonical form of the originalidiom, direct access would fail. However, if the constituent words of the originalidiom have acquired phrase-specific idiomatic meanings, then the meaning of avariant idiom could be generated relatively quickly via ordinary linguistic pro-cessing.

The phrase-induced polysemy hypothesis can be summarized quite briefly.The constituent words of familiar idioms acquire, through repeated use in idiomcontexts, the meanings that are appropriate for the idioms in which they appear.After these phrase-specific meanings have been acquired, then idiom variantsthat preserve the relationships among the constituent idiomatic concepts can beunderstood via ordinary linguistic processing. In this way, idiom variants can beunderstood exactly as literal expressions are understood-by accessing context-appropriate constituent word meanings and identifying the syntactic and seman-tic relations among those constituents.

Implications for Idiom Productivity

Compositionality and Communicative Intentions. Not all idioms, of course,involve phrase-induced idiomatic word meanings. Consider, first, phrasal idiomssuch as spill the beans and pop the question. Such idioms can vary in the extentto which they are compositional (Cacciari & Glucksberg, 1991; Gibbs & Nayak,1989; Nunberg, 1978). In fully compositional idioms, the constituent words canbe mapped directly onto their idiomatic referents. In the idiom pop the question,for example, the verb pop and the noun phrase the question can be mappeddirectly onto their respective idiomatic referents "suddenly ask" and "marriageproposal." In general, idioms that permit word-to-referent mapping will expressn-argument predicates, where n > 1, as in spill the beans. In contrast, theconstituent words in less compositional idioms such as kick the bucket cannot bemapped individually in a one-to-one fashion to the idiom's meaning of "die";only the phrase as a whole can be mapped. Because it is impossible to have aone-to-one mapping from multiple constituents to a single-argument predicate,such idiom constituents would not develop phrase-induced idiomatic meanings.Accordingly, such idiom types generally will not be used in variant forms.Gibbs, Nayak, and Cutting (1989) reported data consistent with this argument.Less compositional idioms are less flexible both lexically and syntactically thanare more compositional ones (see also Gibbs & Nayak, 1989).

Compositionality, however, is neither a necessary nor a sufficient conditionfor an idiom to be varied productively because idioms can be productive evenwhen their constituent words do not acquire phrase-specific idiomatic meanings.Consider idioms that express single-argument predicates, such as two left feet toexpress clumsiness. The noun phrase two left feet contains three constituent

13

14 GLUCKSBERG

words, and these three words, perforce cannot be mapped individually onto theidiomatic referent "clumsy." These words therefore will not acquire phrase-specific idiomatic meanings. Nevertheless, idioms of this type still can be pro-ductive because the semantics of the phrase itself can have direct functionalrelations with the idiom's stipulated meaning. Variations of an idiom will beproductive if the variation plausibly exploits such relations. The phrase two leftfeet, for example, alludes to the grace (or lack of same) with which someonemight dance if he or she did in fact have two left feet, hence the general idiomaticmeaning of clumsiness. Changing the quantifier from two to three simply modi-fies the degree of clumsiness. If two le/tfeet implies clumsy, then three leftfeet,by ordinary discourse processes, implies more than usual clumsiness.

When an idiom's constituents bear functional relations to the idiom's mean-ing, then operations such as quantification, antonymy, and negation will beproductive provided that a plausible communicative intent can be inferred (Cac-ciari & Glucksberg, 1991). The change from plural to singular in Donald Bar-thelme's essay "convicted minimalist spills bean," for example, is productivebecause of the relation between the singular form of the noun bean and theconcept of minimalism. Similarly, the phrase popped the question would nor-mally be difficult to interpret because one usually proposes marriage to only oneperson at a time. If, however, the question popper were a notorious bigamist,then pluralization of the word question might make sense. The constraints ofgeneral world knowledge, together with rules of discourse and conversation,seem as important for idiom flexibility and productivity as are more formallinguistic factors such as compositionality.

Why Spilling the Peas Is Like Kicking the Pail. We are now in a position tospeculate about why certain lexical substitutions seem to be productive, whereasothers are not. Consider, first, the circumstances under which a speaker mightuse a variant form of a familiar idiom. Least interesting are those occasions whena speaker simply misspeaks, as when a colleague recently complained that "re-search grants were becoming scarce as pig's teeth!" The speaker obviouslyintended the meaning of the original idiom (scarce as hen's teeth) and did notnotice his error. More interesting, the listeners in the room at the time understoodthe intended meaning, and some did not even notice the error, even though thesubstitution of pig for hen was contradictory. After all, pigs have teeth, so sayingthat something is as scarce as pig's teeth is to say that it is not scarce at all.Nevertheless, because the phrase was immediately recognizable as a misspokenform of the original (perhaps because of the close semantic relation between pigsand hens as barnyard animals), the intended and contextually appropriate idi-omatic meaning of scarcity came through. Normally, unintentional lexical sub-stitutions are semantically related to the correct or intended word (Fromkin,1971), as when, for example, a speaker substitutes swallow the bullet for bite thebullet.


In each of these examples, a lexical substitution was made unintentionally, butbecause the original idiom was called to mind, the utterances were understoodappropriately in context. More interesting are those cases where a speaker fullyintends the variant form. What kinds of variations do speakers (or writers) use?Perhaps this question might be better if posed in the negative: What kinds ofvariations rarely, if ever, are seen? It is difficult to imagine a context in whichsomeone would choose to say "kick the pail" instead of "kick the bucket" torefer to someone's death. It is also difficult to imagine a context in whichsomeone would say "spill the peas" instead of "spill the beans" to refer tosecrets being divulged. In the pail-bucket example, the original idiom is noncom-positional and opaque, whereas in the peas-beans example, the original idiom iscompositional and transparent. Nevertheless, both variants seem distinctly odd,most likely because no motivation for the lexical substitutions is apparent. Whatcommunicative intent might lead a speaker to choose pail over bucket or peasover beans?

In cases where no communicative intent can be inferred, listeners either maybe totally confused, or might be reminded of the original idiom and simplyaccept the variant utterance as a mistake on the speaker's part (especially if thesituation and context were appropriate to the original idiom). If listeners doaccept the variant as a synonym of the original, then the idiom is said to belexically flexible. In lexically flexible idioms, near-synonyms may be used inplace of the original words, often inadvertently, without making the idiom unre-cognizable. In contrast to accidental variations, speakers intentionally will createnovel idiom forms by using words that bear an interpretable relation to theoriginal, as in pour the beans to communicate that someone were divulgingsecrets quite lavishly. If a listener decides that a speaker's choice of pour overspill is intentional, then the listener might interpret the varied idiom as denoting amore vigorous and egregious disclosure of information than usual. If, however,the choice of pour were to be perceived as a mistake, then no such commu-nicative intent would be inferred.

Speakers, then, seem to produce idiom variants in two ways: inadvertently,with no communicative intent, and deliberately, to communicate an intendedmodification of an original idiom's conventional meaning. The former caseinvolves lexical flexibility, and this seems to be governed primarily by the rela-tive recognizability of an utterance as a variant of the original idiom. Thus, inphrasal idioms that contain both a verb and a noun phrase, changing either theverb or the noun has less of an adverse effect than changing both. Few peoplewould recognize boot the pail as a variant of kick the bucket, for example. Themore interesting case, when speakers intend the variant, involves semantic pro-ductivity. Semantic productivity requires an interpretable relation between origi-nal constituents and their substitutes such that a communicative intent can beinferred. Semantic productivity, as we noted earlier, is the ability of people tocreate new idiomatic meanings by changing relevant aspects of an idiom's indi-

16 GLUCKSBERG

vidual elements. In contrast to simple and unmotivated synonym substitutions,semantically productive operations serve communicative functions: They aremotivated by communicative intentions and so they should be informative. Somerelatively simple productive operations have been subsumed under the rubric ofsyntactic or lexical flexibility. Among these are:

1. Adjectival modification, as in "When drugs are involved, it's time tospeak your parental mind."

2. Adverbial modification, as in "I)id he finally speak his mind?"3. Quantification, as in "As a diverse but purposeful group, you should speak

your minds."4. Tense marking, as in "He spoke his mind."

5. All of the above, as in, "The tenants' association finally spoke theircollective minds."

What is noteworthy about this example is not only that an idiom can be seman-tically productive, but that this particular idiom is one of a group of noncomposi-tional (nonanalyzable) idioms used by Gibbs, Nayak, and Cutting (1989). Recallthat noncompositional idioms should tend to be both lexically and syntacticallyfrozen, yet this idiom seems to be quite productive.

This example suggests that semantic productivity can be independent of bothsyntactic and lexical flexibility, and it may be independent of compositionality aswell. Instead, semantic productivity seems to be governed by the same principlesthat govern such discourse phenomena as adjectival and adverbial modification,quantification, and negation. For example, one can sing sweetly if one actuallywere singing a song, but one cannot sing sweetly if one is singing to the police.In this latter context, singing refers not to a musical activity but to an act ofdivulging incriminating information about others. Thus, even though one cannotsing sweetly to the police, one still can sing like a canary, that is, sing volublyand with unseemly verve and enthusiasm. Notably, there is nothing in the lexicalor syntactic form of the idiomatic verb to sing that constrains adjectival modifica-tion, only the notion of what it means to "sing" to the authorities. Similarly,there is nothing in the lexical or syntactic form of speak your mind that constrainsany of its variants, other than the communicative intentions that can motivate aspeaker to use a variant.

TOWARD A FUNCTIONAL TYPOLOGY OF IDIOMS

Compositionality and Transparency

We begin with the assumption that all idioms, regardless of their degree ofcompositionality, are automatically processed linguistically, that is, lexically,


syntactically, and semantically. For compositional idioms, the results of lin-guistic analysis will be consistent with the stipulated idiomatic meaning, es-pecially when phrase-induced polysemy has developed. In such cases, stipulatedand linguistic meanings coincide, and so comprehension should be facilitated.For noncompositional idioms, stipulated and linguistic meanings will conflict,and so comprehension should be more difficult. The evidence favors this initialclassification: Compositional idioms are understood more quickly than are non-compositional idioms (Gibbs, Nayak, & Cutting, 1989).

For compositional idioms, some relationship between an idiom's componentwords and its stipulated meaning can be discerned. In such idioms, the particularrelationship, together with pragmatic considerations, will determine how an idi-om may be understood and used. There are at least three ways that word mean-ings can map onto the stipulated meaning of an idiom (Cacciari & Glucksberg,1991). The first type of mapping involves compositional idioms that are opaque.In this idiom type, the relations between an idioms' elements and the idiommeaning are not apparent, but the meanings of individual words nevertheless canconstrain both interpretation and use. For the idiom kick the bucket, for example,the semantics of the verb to kick constrain both interpretation and discourseproductivity (see aforementioned examples). Such idioms were dubbed composi-tional-opaque (Type CO) by Cacciari and Glucksberg (1991).3

A second idiom type consists of idioms that are compositional and trans-parent, Type CT. In these idioms, there are one-to-one semantic relations be-tween the idiom's words and components of the idiom's meaning, usually be-cause of metaphorical correspondences between an idiom's words andcomponents of the idiom's meaning. In the idiom break the ice, for example, theword break corresponds to the idiomatic sense of changing a mood or feeling,and the word ice corresponds to the idiomatic sense of social tension. Similarly,the elements of the idiom spill the beans map onto the components of the idiom'smeaning. Spill corresponds to the act of revealing or letting out, and beanscorresponds to the material that heretofore had been concealed or otherwiseunknown. Furthermore, in such idioms the words themselves have acquiredindividual idiomatic meanings, so that one sense of spill in the context of theidiom is "to reveal," and one sense of beans in this context is "information."Included in this class of idioms are both the normally and abnormally decom-posable idioms of Nunberg's (1978) and Gibbs, Nayak, and Cutting's (1989)classification. According to Gibbs, Nayak, and Cutting, normally decomposableidioms have quasi-literal relations between elements and meanings, as in pop thequestion, whereas in abnormally decomposable idioms these relations are more

3 Cacciari and Glucksberg (1991) and others have used the term analyzable instead of composi-tional in their discussions of idiom types. I prefer the term compositional because it connotescomprehension and use. In contrast, the term analyzable suggests a judgmental operation, notinteractive language use.

18 GLUCKSBERG

or less metaphorical, as in spill the beans or break the ice. We see no compellingreason to distinguish between these two subtypes, because phrase-induced poly-semy should develop regardless of the initial kind of relations between an idiom'scomponent words and that idiom's component idiomatic meanings.

The third type of idiom distinguished by Cacciari and Glucksberg (1991) isType M, quasi-metaphorical. In these idioms the literal referent of the idiom isitself an instance of the idiomatic meaning; for example, giving up the shipis simultaneously an ideal or prototypical exemplar of the act of surrendering anda phrase that can refer to any instance of complete surrender. Other examples ofthis idiom type are carry coals to Newcastle to refer to any instance of bringingsomething to a place that already has a surfeit of that something, count yourchickens before they are hatched to refer to any instance of premature confidencein an outcome, and so forth. Included in this class of idioms are such metonymicphrases as bury the hatchet, where the action of burying a hatchet was once anactual part of the ritual of making peace, but is now used to refer to any instanceof peace making in its entirety.

Quasi-metaphorical idioms convey meaning via their allusional content-theycall to mind a prototypical or stereotypical instance of an entire category ofpeople, events, situations, or actions. These idioms use the same communicativestrategy as do metaphor vehicles in expressions such as "my lawyer was ashark" or "my job is a jail. " In these metaphors, vehicles such as shark or jailallude to ideal exemplars of their metaphoric categories-cutthroat people andconfining, unpleasant situations, respectively-and simultaneously as names forthose categories (Brown, 1958; Glucksberg, in press; Glucksberg & Keysar,1990). These metaphors are used to characterize their referents by assigning themto categories that are diagnostic and often evaluative, as in "Margaret Thatcherwas a bulldozer" (Glucksberg & Keysar, 1990; Keysar & Glucksberg, in press).Quasi-metaphorical idioms function precisely as do such nominative metaphors.They simultaneously refer to an idea exemplar of a concept (e.g., total surrender)while characterizing some event, person, or object as an instance of that concept,as in:

Nick: "I don't know what to do about Henry, he continues to do sopoorly in school."

Alice: "Don't give up the ship. I'm sure he'll do better once he im-proves his study habits."

In this interchange, Alice identifies despair about Henry as an instance of sur-render by using the idiom, don't give up the ship. This is accomplished byimplicitly grouping the two actions into the same category: Despairing of Henryand giving up a ship are analogues of one another, and both are instances of thecategory of actions that constitute total surrender. Total surrender, in turn, isreferred to by the allusion to an ideal exemplar of surrender, giving up a ship.


This rough classification of idioms into three general types suggests somedeterminants of idiom comprehension and use. Comprehension and use shouldbe governed by the functional relations between an idiom's elements and theidiom meaning. Lexical substitutions, syntactic operations, and discourse pro-ductivity should be possible whenever those functional relations are preserved.In addition, there must be some discernible communicative or discourse purposethat is served by using an idiom in some form other than the original. As arguedearlier, a listener or reader must be able to infer a reason for any change. It thusfollows that no typology, whether structural or functional, will be fully sufficient.Both the internal semantics of the idiom and the pragmatics of the discoursecontext will always jointly determine idiom comprehension and use, includingidiom variation. To illustrate this approach, I consider how each idiom type maybe used both in their original and in variant forms.

Idiom Flexibility and Productivity

Idioms vary considerably in the extent to which they may sustain lexical substitu-tions, syntactic operations, and semantic productivity. In general, the more com-positional an idiom, the more likely will it be available for variation of one typeor another, but as we already have seen, compositionality alone is insufficient toconstrain idiom use. Consider a noncompositional idiom such as by and large.Such idioms are considered noncompositional because a semantic and syntacticanalysis of the idiom and its elements fails to produce anything that is relevant tothe idiom's stipulated meaning. Hence, word substitutions should not be accept-able. If someone were to say, for example, "by plus large," most hearers wouldnot recognize the utterance as a misspoken form of by and large. Similarly,saying "by and small" to communicate the opposite of by and large also wouldfail because it, too, would not be recognized as a variant of the original. Theprimary reason for such failures is that the component words of noncomposi-tional idioms do not map onto the idiom's meaning except in the most general ofways.

To the extent that there is any semantic relation between an idiom's elementsand the idiom meaning, some semantic productivity should be possible. Theword large bears some relation to the idiom meaning of "generally," and sorelevant modifications should be possible, as in by and not-so-large in a contextthat would support this qualification. The productive use of negation in thisidiom points up another problem for the view of idioms as purely noncomposi-tional strings. If an idiom is truly an unanalyzable whole, then the scope ofnegation-or more generally, scope of modification-must be limited to theentire string. A negation or an adjective cannot be used to modify a semanticallyempty element or constituent within a string (Cruse, 1986). In some cases,modification of idiom constituents, as in break the proverbial ice can be treatedas a metalinguistic comment on the expression as a whole. Nevertheless, there is

20 GLUCKSBERG

a clear and important theoretical difference between such metalinguistic com-ments and true semantic modification, as in the aforementioned by-and-largeexample, or in such cases as "he broke the really frigid ice," where the conceptof social lack of warmth is intensified, not merely commented on. These exam-ples suggest that the idiom by and large is not purely or completely noncomposi-tional because the semantics of large do bear some functional relation to theidiom's meaning. Thus, this idiom is partially, if minimally, compositional.Indeed, purely noncompositional idioms may not exist at all. To the extent that aconstituent of an idiom may be modified independently of the idiom as a whole,it is compositional and so could be used productively in discourse.

More compositional idioms, whether opaque or transparent, are more flexi-ble, but here too pragmatic considerations will be central. For opaque idiomssuch as kick the bucket, where the idiom's component words have nothing to dowith the idiom's meaning, lexical substitutions will be comprehensible only if theoriginal idiom is called to mind. Even then, no communicative intent could beinferred because there are no interpretable relations between a substituted wordand the original. As discussed previously, variants such as boot the bucket andkick the pail might be recognized as meaning "to die" (Gibbs, Nayak, & Cut-ting, 1989), but people would be at a loss to understand why someone would usethese variants. Neither the substitution of boot for kick nor pail for bucket seemsmotivated by any communicative purpose, and so would not be consideredacceptable, unless used by a nonnative speaker. In this latter case, the usagewould be understood but recognized as a mistake. When near-synonyms aresubstituted for both the verb and the noun, as in boot the pail, then the idiomaticmeaning of "to die" is not recognized (Gibbs, Nayak, & Cutting, 1989). Foropaque idioms, then, lexical substitution by near-synonyms are either not under-stood, or viewed as mistakes.

Syntactic operations for idioms also are constrained primarily by the seman-tics and pragmatics of an idiom's components and idiomatic meaning. Considerhow kick the bucket might be varied syntactically. Kicking is a discrete action,and so even though one can lie dying for a week, one cannot say "he lay kickingthe bucket for a week." One can say, for the same kinds of reasons, "almost,will, can, might, may, should, or didn't kick the bucket . . . " The operation ofsemantic constraints can be seen in to o examples of adjectival modification, oneacceptable, the other not. It would be acceptable to say "he silently kicked thebucket" because both kicking and dying can be accomplished silently. It wouldnot be acceptable to say "he sharply kicked the bucket" because there is no wayclear way to understand how anyone could die "sharply" (cf. Wasow et al.,1983).

The operation of pragmatic constraints is illustrated by the nonacceptability ofthe passive voice for this idiom. People tend to reject "the bucket was kicked byJohn" as a paraphrase of "John kicked the bucket." The communicative role ofthe passive form provides a good reason for not using it for such idioms. Passives


are used to put focus on the object of a clause or sentence, usually when there issome prior topicalization, as in: (a) The woman had just turned the corner whenshe was hit by a car; or (b) What happened to John? He was hit by a truck. Nosuch communicative purpose can be served by topicalizing bucket, and so thepassive form is uninterpretable; that is, the use of the passive would not bemotivated. The general principle we propose is: A syntactic operation on anidiom will be acceptable if and only if it produces a comprehensible difference ininterpretation; that is, a reasonable communicative intention can be inferred. Foropaque idioms, the passive form rarely if ever will be acceptable because thererarely would be any reason to topicalize or focus on a grammatical or logicalobject. Tense markings for opaque idiom, in contrast, would be acceptable andinterpretable provided that those tense markings would make sense for the idi-omatic meaning itself; for example, one can die in the future and so one can alsokick the bucket in the future.

Comprehension and use of transparent idioms, such as break the ice and spillthe beans are governed by the same principles that govern opaque idiom use. Acentral difference, however, is that the elements of transparent idioms can bemapped onto the components of the idiom's meaning. Any operations that (a)respect the semantics of each element, (b) preserve the relationship between anidiom's elements and meaning components, and (c) respect the idiom meaningitself should be acceptable and interpretable provided that a reasonable commu-nicative intent can be inferred. Lexical substitutions should be acceptable if theysatisfy these conditions, and so, as we have seen previously, variants of break theice such as crack the ice, break the frost, and break the chill could be acceptable.In each of these cases, the concept of abrupt breaking is preserved and themetaphorical relation between physical temperature and interpersonalwarmth/coolness is also preserved.

Lexical variants that violate the conditions just specified should be consideredunacceptable. To say crush the ice would be unacceptable, primarily because thekind of metaphorical ice involved in this idiom is not the kind that can becrushed: It is, metaphorically speaking of course, thin and brittle, capable ofbeing cracked or, perhaps, even shattered. This example illustrates again how thesemantics of an idiom's elements can govern idiom use and productivity at thelexical level.

At the syntactic level, the same principles apply: Any syntactic operations thatsatisfy both the semantics and pragmatics of the idiom's elements and the idiom'smeaning would be appropriate-again with the proviso that a communicativepurpose can be inferred by listeners. Accordingly, passive transforms would beacceptable if it would be appropriate to focus on a grammatical object, as in "theice was finally broken" or "despite days of intensive questioning not a bean wasspilled." Note that in this latter expression bean was used in the singular. Plu-ralization operations also will be acceptable if they would be appropriate for theidiom's meaning. Beans in this context can be either singular or plural because

21

22 GLUCKSBERG

secrets can be singular or plural: The pragmatics of the idiom's referent are thegoverning factor. The ice in break the ice cannot be varied in quantity becausethe social tension referred to by the term ice is a singular, momentary state. Inother contexts, of course, both ice and social tensions can be quantified.

Allusional Content: Quasi-Metaphorical Idioms. The insufficiency of corn-positionality as a determiner of idiom use and productivity is illustrated clearlyby quasi-metaphorical idioms. These idioms are fully compositional in that alinguistic analysis yields a completely adequate interpretation. The literal mean-ing of carrying coals to Newcastle, for example, is relevant and intended, eventhough this literal meaning is insufficient. Speakers using such idioms intend theliteral meaning and simultaneously allude to the action of carrying coals toNewcastle as an ideal exemplar of the situation that is the momentary topic ofconversation. Even though the idiom itself is fully compositional, lexical sub-stitution is highly constrained. For example, "carrying wood to Birmingham"communicates nothing even close to the original meaning. When, however, acommunicative intent can be inferred, then well-chosen paraphrases can be effec-tive. A newspaper article once reported on the dismal failures of a nucleargenerating plant at Shoreham, Long Island, that was being converted to conven-tional fuels. The headline read, "Carrying Coals to Shoreham." This headlineserved to remind readers of the original idiom, and familiarity with the Shorehamnuclear dilemma made the innovative idiom's meaning clear.

Within the constraints imposed by the metaphorical nature of the coals-to-Newcastle idiom, lexical substitutions and variants are freely available. The verbto carry can be replaced by another verb so long as the action involved isconsistent with the intended meaning and context. Thus, bringing to or from,sending, selling, offering, can be used when appropriate. Like any metaphor,quasi-metaphorical idioms can be tailored to suit discourse purposes.

The principles that govern syntactic operations in general also apply to quasi-metaphorical idioms. Syntactic operations must be communicatively motivated,and so any changes they make in an idiom's meaning must be interpretable incontext. Consider the passive form. For most quasi-metaphorical idioms, nopurpose would be served by focusing on the grammatical object. For this reason,it would make no sense to say "Newcastle was where the coals were carried to."This constraint, however, is not a general one. There can be metaphorical idiomsthat would make sense in the passive form, as in "after intensive discussionsamong the warring parties, the hatchet was finally buried once and for all." Inthis case, the grammatical object, hatchet, can be the focus of the expression.The applicability of any syntactic operation will be governed by such commu-nicative considerations.

Discourse productivity, as before, also will be governed primarily by pragmat-ic considerations. For example, one can easily imagine a context for the follow-ing interchange:


Ken: Don't worry, I'll cross that bridge when I come to it.Ann: By that time they will have burnt it down!

Here, as in earlier examples, the semantics of an idiom element (in this example,bridge) can be used to generate appropriate conversational responses to theoriginal idiom. While retaining its role as symbol, bridge still can be treated as areal bridge so long as its symbolic function is preserved.

IDIOM MEANINGS: LINGUISTIC, STIPULATED,AND ALLUSIONAL

Integrating Linguistic and Other Meanings

All idioms, by definition, have at least two meanings: the linguistic meaning ofthe words and phrases themselves, and the stipulated, idiomatic meaning. Webegan this essay with an analysis of the direct access hypothesis, which effective-ly denied the role of linguistic meanings in idiom use and comprehension. Wehave argued that even for seemingly unanalyzable idioms such as by and large,linguistic meanings play important roles. For such idioms, as well as for other,more compositional idiom types, linguistic meanings are generated automaticallyand are available for use in the context of discourse. Sometimes, as for relativelynoncompositional idioms such as kick the bucket, the linguistic meanings play noinitial roles and do not contribute to initial interpretation (Gibbs, 1984). Forrelatively compositional idioms such as spill the beans, the component wordshave acquired idiom-specific meanings, and so a linguistic analysis produces aninterpretation that is consistent with the stipulated, idiomatic meaning. In allcases, the linguistic and stipulated meanings of idioms always are generated andmust be integrated within the discourse context to provide an interpretation of thespeaker's intended meaning.

Allusional Content. In addition to linguistic and stipulated meanings, manyidiomatic expressions also have what I call allusional content. Whenever anexpression calls to mind a specific action, event, situation, or person, then use ofthat expression can be said to be an allusion to that action, event, situation, orperson. When speakers make an allusion to something, listeners must infer thespeaker's communicative intent. Citing a proverb, a line of poetry, or a song titleare examples of explicit allusions. The literal meaning of an allusive expressionis always intended, but is never enough to convey the full speaker's intention.Consider, for example, someone reciting these words of a familiar song, "I wanta gal, just like the gal that married dear old Dad." If uttered in the context of awedding ceremony, it might be taken as a comment on the similarity of the brideto the groom's mother. If uttered in the context of a performance of Oedipus Rex,

23

24 GLUCKSBERG

it would be taken as a satiric and perhaps tasteless comment on the Oedipuslegend itself.

Here, as with ordinary phrasal idioms, the literal meaning of an expression isnot irrelevant, but instead is an important element of the speaker's intendedmeaning. Listeners, in constructing interpretations of speakers' intentions, mustintegrate the linguistic, stipulated, and allusional content with the discoursecontext. Because the relevance of any one of these three sources of meaningcannot be determined in advance, one comprehension strategy would be toautomatically compute all the meaning types that are available in any givenutterance: (a) literal, (b) stipulated-idiomatic, and (c) allusional. More often thannot, all three sources of meaning are relevant, and this strategy seems efficient aswell. Just as people do not-indeed cannot-isolate or ignore the meanings ofwords and phrases when engaging in discourse, people cannot ignore thosefamiliar, memorized "chunks" of speech whose meanings derive not from thelanguage itself but from their role in everyday experience. Included in thiscategory of language are all those word strings that people learn, such as movieand book titles, song titles and lyrics, poetry, proverbs, cliches, morals, and soforth. These expressions have explicit a]lusional content. We turn now to a classof idioms that have implicit allusional content: those idioms that reflect underly-ing conceptualizations of complex concepts.

Implicit Allusional Content: AccessingConceptual Knowledge

Many idioms in common use seem tc reflect underlying conceptual metaphors(Nayak & Gibbs, 1990). Anger, for example, can be expressed with such idiomsas blow up, blow one's top, let off steam, cool down, all reflecting the conceptualmetaphor of anger as heat in a pressurized container (Lakoff, 1987; Lakoff &Johnson, 1980). Such idioms are not used interchangeably, but are chosen so asto reflect the particular conceptualization that is appropriate in a given discourse.For example, people preferred to say "flip one's lid" rather than "get hot underthe collar" when a provoking incident was serious, but the reverse when aprovoking incident was fairly mild (Nayak & Gibbs, 1990). Nayak and Gibbsstudied idioms that expressed emotions, and found that idiom choice and in-terpretation were consistent with people's underlying conceptualizations of thoseemotions. These findings provide evidence that people exploit the implicit allu-sions to emotional states that are available in such idioms as play with fire (fordanger), touch bottom (for despair), and shake in one's shoes (for fear).

In essence, idioms such as these are no different from the quasi-metaphoricalidioms discussed earlier. Shaking in one's shoes can be understood as ametonymic allusion to a stereotypically fearful behavior, just as burying thehatchet can be understood as a metonymic allusion to a stereotypical part of apeace-making ceremony. Both of these idioms rely on all three forms of mean-


ing-literal, stipulated, and allusional-for their communicative effectiveness.Speakers apparently recognize, albeit implicitly, that some choices of idioms aremore appropriate in context than others, as whenflip one's lid is chosen over dida slow burn to express a particular kind of anger in a specific situation.

Do speakers use word meanings more generally in their choice of idioms?When, for example, might someone choose to say kick the bucket rather thangive up the ghost to refer to someone's death? One possibility is that the kickidiom might be chosen when the death was perhaps unanticipated and abrupt,and give up when the death had been expected and the dying more or lessgradual. Word and idiom choice as a function of specific idiomatic meaningsmay well be governed in general by the relations among an idiom's componentelements, the stipulated (idiomatic) meaning, and the context of use. Far frombeing unanalyzable wholes, these idioms seem fully operational as linguistic andconceptual entities, even though their linguistic meanings are insufficient forcomplete interpretation.

The precise ways in which the three sources of meaning involved in idiomuse-literal, stipulated, and allusional-are used by speakers have yet to beexamined in detail. The precise ways in which these three kinds of meaning areintegrated in comprehension is as yet not understood. What is clear, however, isthat a focus on only one or two of these kinds of meaning to the exclusion of theothers will fail to yield an adequate account of idiom use and comprehension.More specific criticisms of much of the current psychological, linguistic, andphilosophical work in the area of figurative language could be appropriate here,but then, people in glass houses should not throw stones.

ACKNOWLEDGMENTS

I am grateful for the financial support provided by the National Science Founda-tion, Grants BNS 8519462 and BNS 8819657, and by the Public Health Service,Grant HD25826-01 to Princeton University, and to the Council for InternationalExchange of Scholars for a Fullbright travel grant to C. Cacciari, who contrib-uted in important ways to much of my thinking on figurative language. I alsothank Mary Brown, Kay Deaux, Boaz Keysar, and Matthew McGlone for theirvaluable discussions, comments, and suggestions. Correspondence can be sent toSam Glucksberg, Department of Psychology, Princeton University, Princeton,New Jersey 08544-1010.

REFERENCES

Bobrow, S., & Bell, S. (1973). On catching on to idiomatic expression. Memory & Cognition, 1,343-346.

25

26 GLUCKSBERG

Brown, R. (1958). Words and things. New York: Free Press.Cacciari, C., & Glucksberg, S. (1991). Understanding idiomatic expressions: The contribution of

word meanings. In G. B. Simpson (Ed ), Understanding word and sentence (pp. 217-240).Amsterdam, Netherlands: Elsevier Science Publishers (North Holland).

Cacciari, C., & Tabossi, P. (1988). The comprehension of idioms. Journal of Memory and Lan-guage, 27, 668-683.

Cruse, D. A. (1986). Lexical semantics. New 'York: Cambridge University Press.Fromkin, V. A. (197 1). The non-anomalous nature of anomalous utterances. Language, 47, 27-52.Gibbs, R. W. (1980). Spilling the beans on understanding and memory for idioms in context.

Memory & Cognition, 8, 149-156.Gibbs, R. W. (1984). Literal meaning and psychological theory. Cognitive Science, 8, 275-304.Gibbs, R. W., & Nayak, N. (1989). Psycholinguistic studies on the syntactic behavior of idioms.

Cognitive Psychology, 21, 100-138.Gibbs, R. W., Nayak, N. P., Bolton, J. L., & Keppel, M. E. (1989). Speakers' assumptions about

the lexical flexibility of idioms. Memory & Cognition, 17, 58-68.Gibbs, R. W., Nayak, N., & Cutting, C. (1989). How to kick the bucket and not decompose:

Analyzability and idiom processing. Journal of Memory and Language, 28, 576-593.Glucksberg, S. (in press). How metaphors work. In A. Ortony (Ed.), Metaphor and thought (2nd

ed.). Cambridge, England: Cambridge University Press.Glucksberg, S., & Keysar, B. (1990). Understanding metaphoric comparisons: Beyond similarity.

Psychological Review, 97, 3-18.Greenberg-Concool, N. (1990). Don't throw the beans: A study of a young child's comprehension of

idioms (unpublished research report). Princeton, NJ: Princeton University.Keysar, B., & Glucksberg, S. (in press). Metaphor and communication. Poetics Today.Lakoff, G. (1987). Women, fire and dangerous things. Chicago: University of Chicago Press.Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago: Chicago University Press.McGlone, M. S., Glucksberg, S., & Cacciari, C. (in press). Semantic productivity and idiom

comprehension. Discourse Processing.Nayak, N. P., & Gibbs, R. W. (1990). Conceptual knowledge in the interpretation of idioms.

Journal of Experimental Psychology: Generol, 119, 315-330.Nunberg, G. (1978). The pragmatics of reference. Bloomington: Indiana University Linguistics

Club.Ortony, A., Schallert, D. L., Reynolds, R. E., & Antos, S. J. (1978). Interpreting metaphors and

idioms: Some effects of context on comprehension. Journal of Verbal Learning and VerbalBehavior, 17, 465-477.

Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of ExperimentalPsychology, 18, 643-662.

Swinney, D. A., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journalof Verbal Learning and Verbal Behavior, 18, :523-534.

Tabossi, P. (1988). Accessing lexical ambiguity in different types of sentential contexts. Journal ofMemory and Language, 27, 324-340.

Wasow, T., Sag, I., & Nunberg, G. (1983). Idioms: An interim report. In S. Hattori & K. Inoue(Eds.), Proceedings of the XI1th International Congress of Linguistics, Tokyo.

The Place of Idioms in a2 Literal and MetaphoricalWorld

Cristina CacciariUniversity of Bologna, Italy

Idiomatic expressions belong to the vast family of fixed phrases, cliches,proverbs, indirect speech acts, speech formulas, and so forth, that shares somedegree of conventionalization of meaning yet at the same time differs in semanticas well as syntactic properties. As many authors have noted (Coulmas, 1981;Hoffman & Honeck, 1980; Honeck & Hoffman, 1980; Pollio & Burns, 1977), alarge part of our everyday linguistic repertoire is formed by these "conven-tionalized ways of saying things" (Fillmore, 1978, p. 170), so that the usage ofidiomatic expressions seems to be governed by the following informal rule:"Speak idiomatically unless there is some good reason not to do it" (Searle,1975). Not only is their meaning stipulated, but so also is their correct intonation(cf. Hockett, 1958, on politeness formulas), the context, and the linguistic aswell as interpersonal functions.

Nonetheless, the task of defining what an idiomatic expression is, and how itis acquired and understood, is still a rather difficult and controversial one. To thesame extent, the border between literal and idiomatic expressions, on the onehand, and metaphorical and idiomatic expressions, on the other, is also contro-versial. The difficulty in defining idioms is due to several factors that are exam-ined in several chapters of this book. Among the possible sources of such diffi-culty one might include the fact that, as Michiels (1977) noted, idiomatization isa process. A given structure is not idiomatic once and for all, but graduallyacquires its idiomaticity. Furthermore, idioms are at the same time holistic andanalyzable (Coulmas, 1981): They are composed of freely occurring units whoseglobal meaning cannot be reduced simply to the meanings of these units. But atthe same time, for most idioms, people have strong intuitions as to the rela-tionship between the meaning of the constituent words and the idiomatic referent

27

28 CACCIARI

(Nayak & Gibbs, 1990). Last, idioms are not all created equal. As we see,idiomaticity, as a general notion, has come to include very different linguisticphenomena: from partially frozen constructions where the individual words re-tain roughly the same meaning they have in isolation to constructions whosemeaning is totally different from that of the constituents (Lehrer, 1974); fromphrasal verbs to compounds; from indirect speech acts to formulaic expressions.

THE NOTION OF LITERAL MEANING

Idiomatic language always has been defined by differentiating it from literallanguage, which also has functioned as an anchor point for defining metaphoricallanguage. Therefore, an interesting and crucial question is what kind of notion ofliteral language do researchers implicitly or explicitly assume while definingidioms. We briefly sketch the main approaches to literal meaning before going onto consider the place of idioms in a literal and metaphorical world.

Dascal (1987) argued that the notion of literal meaning so far has beenconsidered so central and unproblematic to become an axiomlike notion in theo-ries of language. In successive waves, philosophers, linguists, and then psychol-ogists have criticized the priority accorded to this notion, up to the point ofquestioning its psychological reality (cf. Gibbs, 1984).

The more common notion of literal language still used in linguistics as well asin psychological literature is the one proposed by Katz and Fodor in 1963 that isbased on "the anonymous letter criterion": A person receives an anonymousletter containing only a single sentence. No specification whatsoever aboutmotives, circumstances, or any contextual information is given. What the ad-dressee will understand is the sentence meaning (its literal meaning), its semanticinterpretation rather than the utterance meaning. The latter in fact requires con-textual information that is totally absent in the anonymous letter case aforemen-tioned. One can consider properly only the sentence meaning of the string as itsliteral meaning.

Evidence coming from different linguistic phenomena (deictics, indirectspeech acts, conversational implicatures, irony, etc.) has been used to questionthe plausibility of such a notion of literal language as well as of the assumption of"zero context" implicit in Katz and Fodor's (1963) proposal. It has been arguedthat their proposal is deficient in many respects because it did not take intoaccount linguistic presuppositions, background knowledge, and more broadlythe role of world knowledge in comprehension processes (Clark & Carlson,1981).

Gibbs (1984, 1988) and Dascal (1987) recently challenged the plausibility ofthe general notion of literal language adopted in psychological research, but theissue has been raised previously by other scholars as well (cf. Clark, 1978, 1979,1983; Clark & Clark, 1977; Glucksberg, 1991; Nunberg, 1978, 1979).

2. THE PLACE OF IDIOMS 29

Gibbs (1984) claimed that the "literal language hypothesis," according towhich sentences possess well-defined literal meanings whose computation is thefirst and necessary step of comprehension processes, is based on a rather un-plausible model of humans' linguistic behavior. According to Gibbs, evidencecoming from psycholinguistic studies on literal and figurative language com-prehension showed that the same processes are at work in the comprehension ofboth types of language. He therefore posited that the meaning of highly conven-tionalized string of words, such as idioms or indirect speech acts, is accesseddirectly without any need to compute their literal meanings.

Although Dascal (1987) also maintained the need for questioning the priorityaccorded to the notion of literal meaning, he did not push things as far as Gibbsdid. In fact, Dascal claimed that parts of the literal meaning of a figurativeexpression could play a role in the construction of its meaning, even withoutbeing present in the final interpretation (e.g., in irony). He therefore proposedthe "moderate literalism hypothesis" according to which the notion of literalmeaning must be broadened in order to include conventional meaning attached tocertain expressions. So, for instance, the figurative meaning of frozen metaphors(such as JOHN IS A LION) must be considered as "literal" because of itsconventionality. The figurative meaning of idioms, indirect speech acts, and soon, are accessed directly precisely because their literal meaning has come toinclude also all conventionalized aspects of meaning that were once metaphoricalor indirect.

One might argue that the notion of conventionality is at the same time toogeneral and vague to be a good candidate as a substitute for the notion of literalmeaning. Furthermore, if all language is by nature conventional, how should onedistinguish between different types of conventionality? In any case, the notion ofconventionality seems to be open to the same criticisms addressed to the notionof literal meaning.

According to Lakoff (1986; but see also Fillmore, 1979), the notion of literallanguage has been defined in terms of an idealized and oversimplified model oflanguage and thought with at least four different and rather unspecified sensesattributed to the word literal:

Literal 1. Conventional literality, that is, ordinary conventional languagecontrasting with poetic language.

Literal 2. Subject matter literality, that is, the language normally used to talkabout some domain.

Literal 3. Nonmetaphorical literality,- that is, directly meaningful languagewhose understanding does not require any borrowing from otherdomains of thought and experience, nor any indirect interventionof metaphor or metonymy.

Literal 4. Truth-conditional literality, that is, language capable of "fitting theworld."

30 CACCIARI

According to Lakoff, it is necessary to acknowledge that the notion of literality istheory-dependent and that differences of stress on one or another of the fourlevels aforementioned are often to be attributed to a need for internal coherenceof a given theory. Lakoff's notion of literal meaning rests on the prior notion ofsemantic autonomy (more or less Level 3) (cf. Lakoff & Turner, 1989). Anexpression in a language is semantically autonomous if it is meaningful com-pletely on its own terms; namely, it does not derive any of its meanings fromconceptual metaphors. One can see easily that such a definition leaves almost notheoretical room for the notion that it claims to define.

Defining literal meaning seems to be risky in that one easily can end up takingfor granted the plausibility of the notion of literal meaning in order to defineliterality, nonetheless one might want to take this risk of such a cercle vicieuxprecisely in order to correspond to a basic intuition of native speakers of alanguage. As Rumelhart noted (1979), the classification of an utterance as towhether it involves literal or metaphorical meanings is analogous to our judgmentas to whether a bit of language is formal or not. Both judgments can reliably bemade while being fully aware that a particular utterance may be more or lessfigurative or more or less literal. Purely figurative and purely literal utterancesare at best "rare," but still we are aware of the existence of, and able to use, sucha distinction. This does not imply, according to Rumelhart, any claim or need topostulate different processing mechanisms or strategies.

IDIOMS AND METAPHORS

Idiomatic expressions are taken as one of the most challenging phenomena fortraditional theories of literal language. In fact, idioms behave at the same time aswords and as larger-than-word units (Nagy, 1978). Even in the most frozen ofidioms, the semantics of the single constituent words seem to be still available(cf. the semantic constraints of the verb kick in kick the bucket, Nunberg, 1978;see also Glucksberg, chapter 2 of this book) even though these idioms behavemuch as words, with their meaning acquired and stored in memory, as for othermemorized string of words (lines of poetry, snatches of songs, etc.). Idiomaticexpressions are thus a good test for alternative notions of literality and converselyof figurativeness.

But idiomatic expressions highlight also the need for moving from a one-dimensional level such as the one represented by the dichotomy of literalityversus nonliterality, toward more complex issues, such as, for instance, the oneimplicit in defining what aspects of the meaning of the constituent words of agiven idiomatic expression enter into the comprehension and the interpretativeprocess, if any. Several relationships can exist between a sentence and its referentaside from its being literal or figurative: A sentence in fact can be literal,metaphorical, vague, indeterminate, anomalous, polysemous, indirect, ambigu-


ous, and so forth. The literal-metaphorical dimension is only but one of many,and also a rather controversial one.

Several authors (Fernando, 1978; Fernando & Flavell, 1981; Lehrer, 1974;Makkai, 1972) have taken the view that idiomaticity is more a matter of degreesthan an "all-or-nothing" property of a string in a given language. This is consid-ered to be generally true for all figurative language (Clark, 1978, 1979; Gibbs,1984; Rumelhart, 1979). It has been suggested, therefore, that literal and figur-ative might be the endpoints of a continuum along which one can situate differenttypes of expressions, according to their level of literality, idiomaticity, or meta-phoricity. But is the decision as to whether to locate a given sentence dependingon some well-established criterion or is it purely subjective? We thus are facedagain with the problem of determining the border between literal, idiomatic, andmetaphorical language.

It the last 20 years, several linguistic tests have been proposed, quite unsuc-cessfully, to assess such a border and identify idioms (Hockett, 1958; Makkai,1972). Among the more important are the following two: (a) test by substitutionor replacement whereby any morpheme replaceable by another of the same classidentifies it as nonidiomatic; and (b) once an expression already has been testedby means of (a) (and has also a noncompositional meaning), expose it to as manytransformational changes as its internal structure will permit. But as Fernandoand Flavell (1981) noted, this attempt to classify idioms in terms of their surfaceform, and to distinguish them from metaphors on this basis, got stranded in a seaof counterexamples and exceptions. Fernando and Flavell argued that the onlypossible distinctive criterion is a pragmatic one, namely the one established bythe concrete usage of idiomatic expressions by native speakers that is groundedin their intuitions on language.

Cruse (1986) proposed to distinguish between literal, idiomatic, and meta-phorical expressions on the basis of their relative degree of semantic transparen-cy (i.e., the degree to which an expression is divisible into semantic constituents)or of opaqueness. The degree of opaqueness is definable in terms of two compo-nents:

1. The extent to which the constituents of an opaque expression are "fullsemantic indicators" (i.e., their meaning contributes to the global sense of theword): Blackbird is therefore less opaque than ladybird because it has two fullindicators whereas the latter has only one (bird). They are in turns less opaquethan red herring or in a brown study, which contain no indicators at all.

2. The discrepancy between the combined semantics of the indicators and theoverall meaning of the idiom.

Bound collocations (i.e., sequences of lexical items that habitually co-occur)such as "torrential rain" or "heavy drinker" fell outside this typology because,according to Cruse, each of the lexical items composing them at the same time

32 CACCIARI

preserves its semantics. Accordingly. dead metaphors are not to be considered asidioms because the hearer can still trace the metaphorical origin, even though themetaphorical meaning is possibly encoded by the hearer as one of the standardsenses of the expression (cf. Lakoff & Turner, 1989).

It might be worth noting that Ruwet (1983) argued that one of the possibleways of escaping from the theoretical swamp of the discussion on idioms is tostudy them in status nascenti, namely tc study the figurative origin of idioms (beit metaphorical or metonymical).

The life cycle that goes from a metaphor to an idiom has been traced by Hobbs(1979) in the following way. In the first phase a metaphor is creative and alive: Aterm belonging to a conceptual domain is extended to another domain and in-ferential paths allow it to be interpreted. In the second phase the metaphor isalready familiar and the interpretive path starts to be established and thereforeless complex. In the third phase the metaphor is already "tired"; a direct link isformed between the two domains. In the fourth and final phase the metaphor is adead one. One no longer can trace the metaphorical origin of the expression. Theexpression is conceived of as a way of denoting the object, action, or event thatonce was connotated only metaphorically as such (e.g., the case of "set avariable to a value" or the use of space for designing time expressions).

The relevance of the figurative genesis of idiomatic expressions for a theory ofidioms has been stressed by many linguists and psychologists. A still disputedissue however is the scope of such an interest: (a) Are we simply interested inknowing how it came to be that "die" has been conceptualized in terms ofkicking and buckets?; (b) do we want to understand the movement of the mean-ing from the concreteness of literal language to the abstractness of idiomaticity(Fernando, 1978; Fernando & Flavell, 1981)?; or (c) do we believe that for mostidioms the meanings of the constituent words and their internal structure (meta-phorical, metonymical, based on analogies, and the like) are available and stillplay a role in understanding and interpreting idioms?

A new wave of studies on idioms shares the idea that the latter is the case:There is already evidence enough to refute the traditional idea that idioms thatonce had metaphorical origins have lost their metaphoricity over time and arenow fixed expressions directly stipulated in the lexicon with no possible use ofthe semantics of the words composing them. Several studies (Gibbs, Nayak,Bolton, & Keppel, 1988; Gibbs, Nayak, & Cutting, 1989; Nayak & Gibbs, 1990)were aimed at demonstrating that speakers have reliable intuitions about theanalyzability of idioms, and that these intuitions do play an important role indetermining certain syntactic behavior of idioms, their ease of comprehension,and the extent to which idioms can be semantically productive. One therefore canconsider some classes of idioms as motivated string of words and such moti-vation, far from being a simple etymological fancy, is available and usable incomprehension and interpretation. Before trying to offer more arguments andevidence in favor of this view, the "canon" that has based the standard definition


of idioms (i.e., their being noncompositional lexical units) has to be considered,partly to see if anything of the standard view still holds true or can be integratedin a revised form into a theory of idiom representation and comprehension. Afterall, we certainly do not want to throw away the baby with the bath water.

IDIOMS AS NONCOMPOSITIONAL UNITS

The issue of compositionality of idioms has figured prominently in linguistic andpsychological discussions of idiomaticity. Defining an idiom as an expressionwhose meaning is not formed by the composition of the meanings of the constitu-ent words, according to the morpho-syntactic rules of the language (Katz, 1973),has seemed for years the more straightforward way of defining it. The pointssupporting this claim were of three sorts and we briefly review them.

First, although the semantic interpretation of expressions such as "he brokethe cup" only requires the knowledge of the meanings of the constituent wordsand of morpho-syntactic rules, it seems not to be so for the figurative meaning ofbreak the ice. Such figurative meaning must, according to the noncomposi-tionality hypothesis, be retrieved from lexical memory.

Second, the lack of compositionality represents for some authors (e.g., Cruse,1986) one of the few criteria for distinguishing between idioms and metaphors.In fact, according to this proposal, whereas the metaphor exploits the meaningsof the words in order to induce the hearer to see a thing, a state of affairs, or evenin terms of something else, the constituent words of an idiom are semanticallyempty.

This claim mirrors the opposition between metaphors that are creative (andexploit the "connotative penumbra," Levinson, 1983, p. 150, of the words) andidioms that have fixed and frozen meanings. This opposition between creativityand fixedness is due to the fact that in "figurative formulas" (such as proverbs,idioms, cliches) the metaphorical structure or more generally the "history" thatonce motivated them is now unperceivable. At best, idioms might be conceivedof as frozen or dead metaphors. But as Tannen (1989) recently argued, most ofour speech is prepatterned, which does not amount to saying that language isfixed and "tired." She argued that in the prepatterning of language (includingwhat is usually called idiomaticity or formulaicity), there is a pragmaticallymotivated interplay between fixity and novelty, between "echos" of alreadyspoken words (as Bachtin and Barthes stated many years ago) and simple repeti-tion*. This interplay has a major role in everyday conversation and is a resource ofcreativity.

The third argument used to sustain the noncompositionality of idiomatic ex-pressions is their high degree of internal cohesiveness (or frozenness), a conse-quence of their being multiword lexical units (Fraser, 1970; Swinney & Cutler,1979). So, just as one cannot modify internally a word without changing its

34 CACCIARI

meaning (aside from morphological modifications, of course), the same holdstrue for idioms (Chafe, 1970; Cruse, 1986). For instance, any adjectival modifi-cation of an idiomatic expression will produce either a switch toward the literalmeaning of the string (if any) or a metalinguistic comment on the idiom as a unit(e.g., she or he kicked the proverbial bucket), where the scope of the modifica-tion ranges over the entire string and not, as in literal language, only over theconstituent to the right of the adjective.

Before briefly considering the arguments raised against the idea of a strictnoncompositionality of idioms, it might be worth noting that idioms are sointernally heterogeneous that the generalization of one characteristic (e.g., non-compositionality or syntactic frozenness) to the entire class of expressions can behighly problematic. As Davies (1982/1983) noted, a definition based only onsemantic noncompositionality does not uniquely identify idioms with respect, forinstance, to compound words: There ale, in fact, other expressions that can bedefined as "syntactically complex expressions whose meanings are not deter-mined by the semantic properties of the constituents and modes of combination"(p. 71), for instance, combinations such as "carpet sweeper" and "vacuumcleaner." In both cases, in fact, one cannot work out the meaning from theconstituents and their modes of combination alone: A carpet sweeper sweepscarpets but a vacuum cleaner does not clean vacuums; instead, it cleans by meansof a vacuum (cf. Lehnert, 1986, for a detailed analysis of nominal compounds).A more complex definition of idioms, taking into account at the same time bothcases of noncompositionality and of semantic transparency, seems necessary inorder to uniquely define idiomatic expressions.

SOME CRACKS ON AN "AXIOMATIC IDEA"

First linguists (Gaatone, 1981; Nunberg, 1979; Ruwet, 1983; Wasow, Sag, &Nunberg, 1983) then psychologists (Cacciari & Glucksberg, 1990; Gibbs,Nayak, Bolton, & Keppel, 1988; Glucksberg, 1991) insisted that the "axiomaticidea" (Wasow et al., 1983, p. 107) that the meaning of an idiom is not a functionof the meanings of its parts is deficient in several respects (I am concernedmainly with semantic compositionalit)). Essentially this axiom is not able toaccount for (a) the fact that parts of idioms can be modified by means ofadjectives (e.g., leave no legal stone unturned where the scope of the adjectivedoes not range over the entire idiom but only locally); (b) quantification (e.g.,pull a string or two) and focusing effects (e.g., those strings he wouldn't pullforyou) seem to work to the same extent as for literal language; and (c) parts ofidioms can be omitted in elliptical constructions (e.g., my goose is cooked, butyours isn't).

Wasow et al. (1983) qualified their break with the previous linguistic literatureby claiming that parts of an idiom have typically identifiable meanings that


combine to produce the figurative meaning as a whole. The idiomatic meaning isderived from the literal meaning of the constituent words in conventionalized andnot totally arbitrary ways through meaning transfer mechanisms that they do notspecify fully.

Nunberg (1979) stated that the failure to explain idioms purely on the basis ofthe syntactic fonn or of the content demonstrated that their behavior must beaccounted for on the basis of the relation between these two levels (cf. alsoRuwet, 1983). Nunberg's claim is that these relations are unperceivable if onestill believes that phrasal idioms possess a single semantic representation gener-ally unrelated to the meanings of their parts. The first step is therefore to recog-nize that idioms can be partially analyzable, namely to acknowledge that thereare idioms for which we can perceive a still vivid synchronic relation to the literalmeaning.

As work by Gibbs and collaborators demonstrated, speakers do have assump-tions about the rationale whereby pull strings or kick the bucket mean what theymean, that is, a rationale that one can define as the motivation for an idiom. Notonly can this rationale affect the way in which an idiom is used, but also it canaffect its syntactic behavior (Gibbs & Gonzales, 1985). Of course not all idiomsbehave the same, therefore Nunberg (1979) distinguished among (a) normallydecomposable idioms that are analyzable and imply conventions whereby each ofthe idiom's constituents can be used to refer to the idiomatic referent (e.g., popthe question, keep the pot boiling, break the ice, etc.); (b) abnormally decom-posable idioms where we are able to identify this relation only by virtue ofconventional metaphors that govern the mapping from constituents to idiomaticmeanings (e.g., hit the ceiling, hit the panic button, throw in the sponge, etc.);and (c) nondecomposable, that is, opaque, idioms (cf. Gibbs, Nayak, & Cutting,t989).

THE NOTION OF MOTIVATION IN IDIOMS

The notion of motivation partly overlaps with similar notions that have been usedby linguists and psychologists who adhere to this new view of idioms. It mayprove helpful to try to specify them. An idiom can be semantically compositionalin the sense that its meaning is entirely derivable given the knowledge of themeanings of the constituent words and an ability to detect and use as an inferen-tial tool the rhetorical structure underlying an idiom (e.g., analogical relations,metonymical relations, etc.; we return to this issue later). The analyzability of anidiom is the extent to which a speaker of the language can trace the relationsbetween the two levels of meaning (literal-local and figurative-global). A fairlydifferent notion is that of predictability. According to Nunberg (1978), a regularword use is predictable for a given speaker if she or he believes that it is rational,namely, is following the normal conventions governing its use and beliefs. A

36 CACCIARI

purely conventionalized use of a word has no regularity and therefore it is notpredictable, although there can be different levels of motivation. In this sense,one hardly can predict the rationale whereby, for instance, a certain noun phrase(NP) was selected to be an idiomatic vehicle. One can speculate on its generalsemantic features that can make it a good candidate for a certain figurativereferent, as in, for example, the case of spill the beans. Beans are countable,numerable, discrete, and so on; therefore they can be taken to represent secretsmuch more efficiently than mud, which is a mass noun. But the reasons whybeans were selected instead of pebbles is rather opaque, although it has beenargued that symbolization processes are not totally arbitrary and are culturallybound (cf. for instance Holland & Quinn, 1987).

The notion of motivation has a long history: Ullmann (1962), for instance,claimed that any language has both conventional and motivated terms. It is asdifficult to imagine a language with no onomatopoeic terms or transparent meta-phors as it is to imagine a language with only motivated terms. There are at leastthree possible levels of motivation: the phonetic motivation (e.g., "boom"), themorphological motivation (e.g., "retell" or "thinker," where the knowledge ofmorphological rules will guide the interpretation), and the semantic motivation,as in figurative expressions (e.g., "the turning point of the question"). Bothmorphological and semantic motivations contain a level of arbitrariness: In fact,whereas affixes are motivated, the components are as arbitrary as the relationshipbetween the words and the objects designated by the metaphor of the turningpoint.

Recently, Lakoff (1987) and Gibbs and collaborators (Gibbs & Nayak, 1991;Nayak & Gibbs, 1990) argued that idioms are not dead metaphors with arbitrarilydetermined meanings. The figurative meanings of idioms are motivated byspeakers' tacit knowledge of the conceptual metaphors underlying them (e.g.,THE MIND IS A CONTAINER or LIFE IS A JOURNEY). These conceptualmetaphors form part of the link between an idiom and its figurative meaning andconstrain the way people understand them, or how people form mental images(Gibbs & O'Brien, 1990). These mappings of source and target domains existindependently in the conceptual system and function, according to the propo-nents of cognitive linguistics, to structure and organize many aspects of ourthought and reasoning. According to this view, metaphorical motivation of idi-oms is rooted at the level of these conceptual structures, and not in a correlationbetween word meanings: The motivation is thus at the conceptual, not at thelexical level.

Although interesting, the idea of conceptual templates shaping all aspects oflanguage production and interpretation seems not to take fully into accountcertain important aspects of language use. First, words do have meanings that arecomputed automatically and cannot be simply bypassed. Second, words aregrouped together in idiomatic expressions most often not arbitrarily but accord-ing to principles that are per se meaningful (use of pars-pro-toto principles,


analogies, extensions of prototypical cases, etc.). Third, people can interpretlanguage sensibly at that level, and do so in a creative way (see later). So it is notsimply a matter of componential meaning (as it was according to standard theo-ries of idioms), but of the interaction between all the meanings (be they central orperipheral ones) that are carried by the words composing the idiom and thestructure underlying the idioms. Although idioms possess established meaningsthat are easy to conceive of in terms of stored semantic concepts, they simul-taneously exploit the senses of the words, the structure underlying their groupingin an idiom, and possibly also conceptual templates such as the ones postulatedby Lakoff and Gibbs (cf. Gibbs & Nayak, 1991; Kreutz & Graesser, 1991).

THE ROLE OF THE CONSTITUENT WORDSIN IDIOMS' INTERPRETATION

One might argue that we now are faced with another cercle vicieux: How can wedistinguish different degrees of transparency of an idiom when we already knowits meaning? Reagan (1987) argued that when faced with the problem of seman-tic decomposability of idioms, one has to distinguish between breaking downmeaning into parts and building up meaning from parts. These two mechanismsshould allow different predictions as to the level of semantic transparency ofidioms. As Ruwet (1983) noted, the confusion between acquisition and in-terpretation has favored the idea of semantic opaqueness of idioms' constituents.

Coulmas (1981) considered the case of non-native speakers faced with theproblem of making sense of idioms as a heuristic model. A possible and sensibleassumption a non-native speaker might make is that any incomprehensible ex-pression she or he might run across could be an idiom. He or she can use threepossible sources of knowledge in order to reduce the range of possible meanings:the logic of everyday activity, because most communicative acts take place instereotypical social situations; the structure of the lexical phrase; and theawareness of the fact that most languages have "idiom-prone lexemes," as take,put, and so forth. By using together all these sources, a non-native speaker mightarrive at the idiomatic meaning "by inference" rather than "by instruction." Thisknowledge, coupled with an attention toward context, can help him or her toarrive at the meaning, or at least at the semantic domain to which the idiombelongs, although it will not be sufficient to tell when and how to use it appropri-ately.

There are several ways that one can investigate the role of the meanings of thewords that form an idiomatic expression. Cacciari and Glucksberg (1990) (seealso Glucksberg, McGlone, and Cacciari, in press, and Glucksberg, chapter 2 ofthis volume), for instance, studied the semantic productivity of idioms, namelythe ability of people to understand and produce new idiomatic meanings bychanging, in a pragmatically sensible way, various aspects of an idiom's indi-

38 CACCIARI

vidual elements (e.g., burn the candle at three ends instead of burn the candle atboth ends). They showed that, according to the principles that govern discourseprocessing, idioms are analyzed both semantically and syntactically. The "literalmeanings" of the constituent words are available and used when necessary, thatis, especially when idioms are analyzable and direct, one-to-one mappings fromidiomatic constituents to idiomatic referents are possible. The words composingsuch idioms can acquire new senses, that is, the ones they get from being part ofan idiom. So spill can come to acquire a "divulge" sense, even though not as adominant one, precisely because of the repeated usage of the idiom spill thebeans together with some level of conventionalization of a figurative and partlymotivated origin. Therefore its meanings will be available and activated accord-ing to different processing needs, types of contexts, tasks, and so on.

Additional sources of evidence for this role of the meanings of the constituentsare represented by production phenomena such as slips of the tongue, where asemantically related word is substituted (e.g., swallow the bullet instead of bitethe bullet; cf. Cacciari & Glucksberg, 1990; Fromkin, 1971); advertising inmedia, and intentional switches between literal and figurative senses of words asthe ones exemplified in the following passages taken from Broderick (1975): (a)"Baldwin occasionally stumbles over the truth, but he always hastily pickshimself up and hurries on as if nothing had happened" (Winston Churchill; p. 8);and (b) "Kennedy won by an eyelash and some thought it was a false eyelash"(Howard K. Smith; p. 9). Broderick interpreted these two examples as caseswhere the literalization of some words (e.g., stumble over) resurrects the meta-phor. The two passages illustrate the possibility of multiple semantic processing,namely the simultaneous processing of the semantics of an idiom and of itsliteralization, at the same time accounting for metaphorical processes at work inthe literalization.

But it seems also possible to interpret the two passages in a different way. Thetwo passages play exactly on the overlapping of the literal semantics of thewords, that is, the ones they possess if taken outside the idiom, as well asthe aspect of meaning they acquire when embedded in figurative expressions. Itis precisely because the meanings are still available that one can produce such aneffect: It is a movement that backgrounds and foregrounds the literal and thefigurative meanings of the words, that is, pushes them on stage and back to thebackstage, according to the communicative needs of the moment. A movementnot dissimilar from the one that we experience when faced with perceptuallyambiguous stimuli: Once we see a vase, but a moment later we see two profiles.The two images can even co-occur for a while; in any case we are aware of theircoexistence as we are aware of the existence and interplay of the two meanings:the literal and the figurative.

As we mentioned earlier, the idea that idioms still have an internal semanticsthat can be used per se in order to allow people to interpret them (the so-calledtransparency of the figurative meaning) might lead to a chicken-or-egg critique:


Do we arrive at the figurative meaning and then have intuitions about the idiom'sanalyzability because we already know its conventionalized meaning (the "breakdown hypothesis" made by Reagan, 1987) or do we arrive directly at such afigurative meaning building it up from its "building blocks" (as for metaphorsthat do not have a stipulated meaning)?

The idea of a continuum from opacity to transparency leads us to hypothesizethe existence of four possible types of idioms: (1) totally opaque idioms; (b)retrospectively transparent idioms: namely, transparent once you either know themeaning (and this allows one to trace the correspondence between literal constit-uents and figurative referents) or are reminded of the episode or setting thatoriginated the idiom; (c) directly transparent idioms: the senses of the wordsleading one to the idiomatic meaning of the string, perhaps by means of the re-creation of an analogical or metaphorical mapping; (d) figuratively transparentidioms: idioms composed of other idioms, or parts that appear in other idioms oras metaphorical vehicles.

What remains nonetheless unclear is whether the idea of such a continuumbears any correspondence to the interpretative strategies used in idiomatic lan-guage interpretation or whether it merely represents an exercise in the stylisticproperties of language. As it might prove useful for most chicken-or-egg prob-lems, it is worth trying to look at it from different perspectives and gatherevidence to try to transform it into a more manageable issue. The preliminaryresults of some experiments on adults and children might shed some light on theissue of idiom transparency and interpretation.

COPING WITH UNKNOWN IDIOMS:STRATEGIES FOR INTERPRETATIONS

The first experiment was conducted on 32 adults.1 They were given a writtenquestionnaire containing a list of unfamiliar idioms (each embedded in a shortsentence containing a personal pronoun + idiom, e.g., "He made the feathersfly") taken from a study on idiom familiarity (Schweigert, 1986). The experi-mental materials consisted of 48 idioms divided into four lists (see Appendix A),and subjects were assigned randomly to one of them. The subjects' task wastwofold: They first were requested to give a paraphrase and rate their confidenceon it (on a 7-point scale going from I :very unsure to 7:very sure) for each of theidioms whose meaning they thought they knew. For those that they were uncer-

'This pilot study was conducted in collaboration with Dr. Boaz Keisar (currently at the Universityof Chicago) and Sachi Kumon (Princeton University) while I was a Fullbright visiting scholar at theDepartment of Psychology of the University of Princeton. Eleven subjects were students at thePrinceton University and 21 were American students at the Johns Hopkins University in Bologna,Italy. All served as unpaid volunteers.

40 CACCIARI

tain of or did not know the meaning, they were requested to make a guess andwrite down a paraphrase on a separate side of the questionnaire. Once finishedwith this first task, they were asked to go back through all the guessed idioms andbriefly write under the guessed paraphrase how they arrived at the meanings ofthose idioms, namely "what made you choose the specific interpretation, did youuse something as a cue, were you reminded of something?" The aim of this pilotstudy was to investigate people's ability to make sense of unfamiliar idioms,namely their ability to use different sources of knowledge and strategies in orderto assign a meaning to a string of words.

The responses were discussed and evaluated both separately and collectivelyby the researchers with over 86% agreement. The written self-reports were usedin order to outline the strategies that students used during the guessing phase.Five such strategies were identified. Sometimes students made mistakes: Theywere sure that they knew the meaning of the idiom, and that was not always thecase. Sometimes subjects confused two similar idioms and the like (47.6% of thetotal answers were correct, with an average confidence rate of 6.4; 11.7% wereincorrect, with a confidence rate of 2.1; and 35.2% were guessing; 5.5% wereincomprehensible or missed).

We were not particularly interested in the average error rate, that is, in theperformance per se, nor in which of the strategies was more effective in guessingthe correct meaning. Rather, we were mainly interested in students' self-report ofthe principles they thought governed their interpretation, and in this respect theywere quite cooperative and accurate. I therefore report here only on the fivestrategies we identified. These types of interpretative paths are interesting evenwhen they led to wrong solutions. In fact, as we see, sometimes students endedup with wrong meanings, although the inferential path was sensible.

Type 1. Look for another idiom that is similar to some extent. This generalprinciple can be instantiated in different ways:

1. See whether one part or construction (e.g., one or more constituents) isinterpretable according to other idioms. Examples are the by the board in go bythe board interpreted as "by the book" in "go by the book";feathers of make thefeathersfly interpreted as in "smooth someone's feathers."

2. Search for another idiom having a related word (e.g., an antonymousword) or the same verb in the same position and interpret it accordingly orchange a constituent and interpret the idiom accordingly. Examples are dropinstead of explode in explode a bombshell; or dry behind the ears interpreted as"wet," so "innocent"; take the bit in one's mouth interpreted by inverting theactive and passive poles, therefore as a passive action.

3. See if there is another idiom that uses a semantic associate of one of thewords of the idiom and interpret it accordingly. An example is get out of the hole,which was interpreted according to "put yourself in a hole."


Type 2. Interpret one constituent literally and use also the other wordscomposing its semantic field to infer the meaning of the idiom. Some examplesare: for hose somebody down: Hose = clean, so clean somebody up = defeatsomebody in a game, therefore hose down = defeat; for polish the apple: polish= clean, improve, and apple = perfection, so polish the apple = reach perfec-tion.

Type 3. Look for a cause-effect relationship between the action describedby the idiom and select either the cause or the effect as the meaning of the idiom.Take for example grease the wheels: "greasing the wheels allows the machineryto start moving" so the idiom means "go faster."

Type 4. Consider the semantics of the words composing the idioms and tryto derive a possible meaning for the idiom. This can imply:

1. Consider the literal semantic of one word. Some examples are back in-terpreted as "beginning" or "source" as in back to the salt mines; or hammer inunder the hammer-what is a hammer used for? Breaking things that are hard(e.g., nuts), so the idiom could mean examine very careful and intensely aproblem. Asleep at the switch "has its sense because of interpretation of the wordmeaning. Switch: railroad switching station."

2. Consider the literal meaning of the entire string, and possible outcomes(reconstruction of the analogy). For example, asleep at the switch implies beingnot careful, not alert. For take the bit in one's mouth, "if you relate the idiom toriding horses, its meaning becomes clear." For darken one's door, "the person'sshadow appears on the door when he or she comes to visit."

3. Interpret the words metaphorically. For example, for a straw in the wind,straw is a symbol of weakness, of no direction, "a straw blowing helplessly inthe wind." For the bottom dropped out, "the bottom symbolizes the depression."For darken one's door, "cast a shadow implies darkness, a negative meaning; thedoor stands for one's self, home."

Type 5. Try to visualize an image of the action/state described. For exam-ple, for bow and scrape: "I had a sort of visual interpretation. I pictured some-one acting sort of like a puppy when it begs"; for make the feathersfly: "causingfluster or upset, the visual image of feathers flying."

What the results of this pilot study seem to show is that the ability to makesense of idioms is not restricted to the simple mechanism of retrieving an idiom'smeaning from lexical memory.2 In case of familiar idioms, people certainly

2Fillmore, Kay, and O'Connor (1988) noted that there are idioms (as let alone, or themost . .. the most) that are partly open and must be filled in according to semantic as well aspragmatic constraints. In these cases, the mental lexicon must incorporate different levels of informa-

42 CACCIARI

retrieve the meaning stored in memory, often together with information muchmore complex than a simple paraphrase. In fact, it has been argued that manyidioms have rich semantic descriptions associated with them that might not befully captured by a single word paraphrase (Gibbs & Nayak, 1991; Michiels,1977; Mueller & Gibbs, 1987; Nunberg, 1978). In fact to use an idiom such askick the bucket does not simply mean 'die," but also dying in a certain way. Forexample, one would not use it to refer to a relative, or in any respectful circum-stance or the like. Some idioms also can take slightly but relevant differentnuances of sense depending on the context, showing that their general meaningmust be instantiated according to the available information. In fact, one mightthink of the function of idioms in terms of specification of existing concepts in away that is not already specified in semantic memory by existing lexical items.

There are thus semantic as well pragmatic constraints on idioms' use that gobeyond the scope of a simple paraphrase. These constraints show that peoplepossess much more knowledge of idioms than simple paraphrases. Of course,they actualize and make use of such knowledge and interpretative paths (basedon similarity, analogical transfer, inferences, etc.) only when necessary and thatis obviously the case for unfamiliar idioms (that might be reasonably seen asmetaphors).

It is precisely such complexity that lenders idioms so difficult to manage for anon-native speaker. As Coulmas (198 1 noted, she or he has to learn that in manycases linguistic meaning is vastly more complicated than the simple computationof the meaning components of identifiable parts, and includes also the range ofappropriate circumstances and applications.

But is there anything special about the strategies people use when trying tomake sense of unfamiliar idioms; namely is there anything new or different fromwhat we routinely do during a conversation or while reading a newspaper article?More generally, is there anything special in making sense of figurative languagewith respect to literal language, lexical or sentential innovations, and so on?There is not enough evidence to sustain the idea of a principled differencebetween types of knowledge, strategies, and meaning construal rules used forunderstanding and interpreting one or another type of language (if such a sharpdistinction still holds, which is highly disputable). In fact, first we make use ofthe same types of knowledge (although different priority in accessing one oranother source of knowledge might occur). Second, we spend most of our ac-tivity of listening or reading in filling gaps, drawing inferences not alwaystriggered by conversational implicatures a la Grice, looking for prototypicalcases or similar situations that might function as anchor points, and the like. Asevidence on lexical innovations, eponymes, indirect speech acts, "nonce"

tion, from grammatical patterns, to an entailment relation in the background of a presupposedsemantic scale for idioms such as let alone; in fact the interpretation of any let alone sentence requiresseeing the two parts of the construction as points on a scale."


senses, and complex concepts showed (Clark, 1983; Clark & Gerrig, 1983;Gerrig, 1989; Gerrig & Murphy, 1991; Medin & Shoben, 1988; Murphy, 1988),we have developed a fairly efficient ability to decode and assign meaning andcoherence even to contradictory or incomplete verbal materials. The generalmechanisms for meaning extractions are so powerful and efficient that it isdifficult to accept the idea of something special used only for figurative lan-guage, as most stage models posited (e.g., Searle, 1979). The general ideaunderlying stage models is that figurative meaning understanding depends on therecognition of some sort of contextual deficiency of the sentence, if taken liter-ally. But this has led to predictions that generally have been disconfirmed by theexisting evidence (cf. Gibbs, 1984; Glucksberg, Gildea, & Bookin, 1982).

HOW CHILDREN ASSIGN A MEANINGTO IDIOMS

For many years, the predominant view was that children up to the teenage yearswere unable to understand idiomatic expressions. Recent work on the develop-ment of figurative competence (Ackerman, 1982; Cacciari & Levorato, 1989;Levorato & Cacciari, 1992; Winner, 1988) has cast doubt on this assumption. Agrowing body of evidence now suggests that even 7-year-old children are able tounderstand idiomatic expression, especially when embedded in highly infor-mative contexts. It therefore can be of some interest to see the strategies adoptedby fifth-grade children (10-year-olds) when asked to assign meanings to idi-omatic expressions. This could shed light on both their metalinguistic awarenessand their figurative competence.

As part of a larger set of studies on the developmental trend of the acquisitionand production of idiomatic expression,3 Levorato and I gave a questionnaire to45 Italian fifth graders and a group of 30 university students (I will not report ontheir results here). Three lists of 10 frequent idioms each, being either seman-tically transparent idioms (e.g., cry over spilled milk), quasi-metaphorical idioms(based on a figurative comparison as in to be as two drops of water, which meansto be very similar), and opaque idioms (e.g., break the ice) were prepared (seeAppendix B). Each list was assigned to a group of 15 fifth graders. For eachidiom, the child was requested to write down (a) a paraphrase of the idiommeaning, (b) the reasons motivating the meaning of the idiom, (c) whether a firstgrader (a 6-year-old) might understand it, and (d) what the 6-year-old can do inorder to understand it. The questionnaire was the same for children and adultsbecause the latter acted as baseline. Adults therefore were advised that this was

3For more details, see Cacciari and Levorato (1989), Levorato and Cacciari (1992), Cacciari &Levorato (1991), and chapter 6 by Levorato in this book.

44 CACCIARI

part of a study on children's acquisition of figurative language. Let us see someof the results concerning children's answers.

First of all, as was predicted on the basis of previous studies, 10-year-oldchildren already have a good competence on idiom meaning, especially for thequasi-metaphorical ones (henceforth, QM) (69.9% of correct paraphrases,51.9% for opaque idioms [0] and 47.9% for transparent idioms [TI). Not manychildren wrote that they had no ideas as an answer to the question: "Why do wesay so," namely the reasons motivating the idiom (13.3% of the total answers).The explanations of the correct meaning of the idioms (43.3 for QM, 26.6% forT, and 15.9% for 0) were classified according to the following categories:

1. Usage conditions: "we use it for saying it," "it's a way of speaking," "it'sa proverb."

2. Causal explanations: e.g., in the case offallfrom the clouds (that means"be astounded"): "clouds are far away, so if someone is on the clouds is faraway from reality that's why she or he doesn't realize what is really happening."

3. Explanations based on the literal state or action expressed by the idiom,its impossibility, outcomes, or agents: e.g., in the case of make a hole in thewater ("be unsuccessful"): "it is impossible to make a hole in the water, so youlose time and realize nothing"; or, for drown oneself in a glass of water ("get lostfor nothing"): "it is impossible because a glass is too small, so it means that youare worried for nothing"; or for be as dumb as afish (that means to be silent andkeep secrets): "fishes do not speak."

4. Explanation based on the literal and figurative outcomes of the idiomaticaction: e.g., for touch the sky with afi'nger ("be extremely happy"): "in order totouch the sky one has to make huge jumps that are possible only if one is veryhappy"; for break the ice: "we say like that because the ice is tough and the morethe time is passing and the more the silence gets iced," or "because words glideas the ice, and breaking it is like getting to the point."

5. Explanation based on analogies: e.g., for close shop: "closing the shop islike to stop working, which means that you keep silent and do not speak to otherpeople"; or for look for a needle in a haystack ("do something that is impossi-ble"): "because looking for a needle in a haystack is as much difficult as lookingfor answers to impossible questions"; or for be among the clouds ("be veryabsent-minded"): "because if one is among the clouds she or he is not on theearth; that is there is metaphor: earth stands for reality, clouds stand for fantasy."

6. Explanations based on perceived symbols: e.g., for be at the seventh sky("be in seventh heaven"): "we all know that sky is wonderful, so if there was aseventh one, can you imagine?"; "in ihe heaven everything is happy and nice."

7. Explanations based on other idioms: e.g., for touch the sky with afinger:"because we can also say to be on the stars, because when one is very happy sheor he seems to fly."


Children were requested also to answer a question concerning younger children'sability to understand idioms and a possible way for them to get to the rightmeaning. The rationale of the question was to elicit as much information aspossible on the tacit knowledge children possess on idioms, namely to gatherother evidence on both the perception of a semantic motivation and the strategiesfor interpretation. As a side effect, it revealed the level of metalinguisticawareness of 10-year-old children, and that children indeed possess a naivetheory on younger children's knowledge and reasoning abilities.

First, many children think that first graders can understand idioms (QM:38.6%; T: 43.9%; 0: 40.6%). Their answers to the question "What can she or hedo in order to understand it?" were classified according to the following types:

1. Exemplification (5.7% on the total of the answers).2. Perform the idiomatic action (20.1%).3. Perform the literal action (12.2%).

4. Ask adults (15.5%).5. By reasoning on the constituent words and on the action described by the

idiom (9.2%).6. By a correct similitude (the child either depicts a similitude or relates the

action to her or himself) (11.7%).4

Let us see some examples associated with the three different types of idioms.

Quasi-Metaphorical Idioms. The more preferred answers are: by a sim-ilitude (22.6%) (for be as dumb as afish: "You must say to her or him that fishesdon't speak, but also that you can say the same thing also for human beings";"make him think how can she or he possibly speak under the water"); for feel asafish outside the water: "It is enough to try to imagine to be a fish and then to tryto figure out how can feel a fish without its water" by reasoning on the words andon the action ( 18.6%) ("In order to understand, she or he can reason about it, payattention, or observe the others"; "examine carefully the meaning of thewords"); perform the literal action (16.6%) (for be afraid also by one's ownshadow: "Stand in front of a light bulb and watch at his shadow"; "it is enoughthat it enters his consciousness the idea that phantoms and living shadows do notexist"; for be as dumb as a fish: "She or he goes to the zoo, and touches theaquarium, if she or he doesn't hear any noise, she or he will understand thatfishes are dumb"; for go backwards as crayfishes: "She or he goes to theseaside, looks for a crayfish, and then makes it walk"); to ask adults (15.9%)

4There were two other groups of answers: a group formed by answers such as "I don't know,"incorrect analogy or incorrect idiom (7.3%), and another one formed by answers such as "she or hecannot understand because she or he is too young" (4.6%).

46 CACCIARI

("you don't have to be a genius for understanding it, but first graders are small soan older person must explain the meaning"); perform the idiomatic action (7.9%)(for be as dog and cat; i.e., "enemies": "She or he can try to argue with its bestfriend"); exemplification (5.9%) (for be as two drops of water: "One mustexplain to him that they are as similar as two white sheets used for drawing").

Transparent Idioms. The answers were classified as follows: Perform theidiomatic action (34.6%) (for look for a needle in a haystack: "They too canunderstand, because they too have lost several things that ended up being some-where," or "if she or he looks for a small thing in a huge room, it's impossible tofind it"; for cry over spilled milk: "I'll show them a movie with a man that hasrepented for having done so many bad things"); perform the literal action(18.6%) (for cost an eye of the head. i.e,, "an arm and a leg": "thinking to buysomething and paying for it with an eye"; for make a hole in the water, i.e., "beunsuccessful": "Let him try to make a hole in the water, he'll see it's impossi-ble"; for cry over spilled milk: "One can spill some milk and then becomedesperate"; for be among the clouds, i.e., "be absent-minded": "One can drawthe picture of a child on a cloud that is not thinking of anything"); by similitude(8.6%) (for cry over spilled milk: "I'll tell her or him that as when someone spillsmilk and cannot then pick it up, when one performs badly she or he cannotremedy it"; for cost an eye of the head: "She or he has to think how unhappymust be a person without an eye"); ask adults (7.9%); exemplification (5.9%)(for be among the clouds: "I'll make her or him an example: While the teacher isexplaining the lesson you're thinking of something else"; for be on the thorns,i.e., "be very anxious about something happening": "I'll use the example ofcartoons: When the cartoon is split and you don't know the end"; for makechickens laugh, i.e., "be socially unsuccessful," namely to tell something that isneither amusing nor believable: "She or he can try to make chickens laugh; it'simpossible"). Few children, many less than with QM idioms, quoted the self-reasoning strategy (1.9%), which can imply that children thought T idioms weremore semantically complex than QM idioms.

Opaque Idioms. Although children provided answers for most of the QMand T idioms, there were fewer answers for 0 idioms (21.2% of the total aremissing, vs. 17.2% for T idioms and 1.3% for QM idioms). The answers weredistributed according to the following categories: Ask adults (22.6%); performthe idiomatic action (17.9%) (for cut the rope, i.e., "escape": "She or he shouldtry to be feared and escape very quickly"); by reasoning on the words (7.2%) (forbreak the ice: "she or he can dwell in the word break); exemplification (3%) (forgive oneself some air, i.e., "beast": "She or he can arrive at understanding byreasoning on the fact that the words do not mean take a fan, but something else";for touch the sky with a finger: "if he or she thinks that in order to touch the skywith a finger one has to make extremely high jumps that are possible only when


one is very happy"; "think to the fact that the sky is the happiest place so howcan one be unhappy there?"). Almost absent is the perform-the-literal-actionanswer (1.3%) (forfallfrom the clouds, i.e., "be astounded": "make her or himfall from the bed, and see the reaction"). Table 2.1 summarizes the preferencesaccorded to the different answers.

There was also a group of answers that one can define as "reflections onyounger children's ability" where 10-year-old children declared why youngerones cannot understand a particular idiom of the list (3.9% on the total of QMidioms, 0.7% of T idioms, 9.3% for 0 idioms). Here are some examples for theidioms where such an answer is more frequent, opaque idioms: for cut the rope,i.e., "escape": "because he or she really believes that it really means to cut arope"; for break the ice: "She or he won't understand because listening tosomeone who is breaking the ice she or he will not think to make friends"; fortouch the sky with afinger: "She or he won't understand because touching thesky with a finger doesn't seem to be such a happy thing."

The three types of idioms seem to elicit different responses and interpretativestrategies. One might attribute this to experimental artifacts such as our choice ofthe materials and of the categories as well as the decisions concerning the best fitbetween answers and categories. However, we believe that our data reflect agenuine effect that can be explained on more motivated grounds, namely thatidioms do differ according to their semantic structure and that children aresensitive to such differences. As a consequence, the interpretative strategieschildren thought of as more appropriate reflect the perception of the semanticcharacteristics and cognitive complexities of idioms. Of course, children's an-swers also reflect a difference in the availability of sources of information andexperiential bases.

Let us compare two types of answers: the perform-the-idiomatic-action an-swer and the perform-the-literal-action one. The first prevails in both T idioms(34.6%) and 0 idioms (17.9), although is much more present in T idioms. This

TABLE 2.1Percentages of Answers to the Question

'What Can She or He Do in Order to Understand It?"

Types of Answer OM T 0

Exemplification 5.9% 5.9% 5.3%Perform the Literal Action 16.6% 18.6% 1.3%Perform the Idiomatic Action 7.9% 34.6% 17.9%Ask Adults 15.9% 7.9% 22.6%By Reasoning 18.6% 1.9% 7.2%By a Correct Similitude 22.6% 8.6% 3.9%No Answer 1.3% 17.2% 21.2%Other Answers 6.9% 4.1% 10.9%

48 CACCIARI

result seems coherent with evidence on adults reported by Popiel and McRae(1988) and Cronk (1990), namely that many idioms that possess a possible literalinterpretation are interpreted mostly figuratively because the likeliness of theirliteral usage is very low. So the children who chose this answer may have reliedon the action referred to by the idiomatic meaning as the more salient andplausible, although possibly not for the same reason for the two types of idioms.The perform-the-literal-action answer is used more frequently in T idioms(18.6%) and QM idioms (16.6%) than in 0 idioms (1.3%). In the first case thismight be because the literal meaning of T idioms, by definition, bears a certainsimilarity to its figurative meaning and the constituent words map onto theiridiomatic referent in a quite direct way. Therefore children might have thoughtthat performing the literal action is a good strategy for getting to the figurativemeaning. In fact, such a strategy is virtually absent for 0 idioms (1.3%). Thepreference for this answer in QM idioms (16.6%) can be largely due to theseidioms being built upon a comparison that specifies a certain state, property, oraction with respect to a metaphorical vehicle that seems to represent it in a moreappropriate or prototypical way. Thus the literal information is actively part ofthe idiom's meaning. The fact that "by reasoning" and "by a correct similitude"are used almost only with QM idioms (18.6% and 22.6%, respectively) can bedue to the structural characteristics aforementioned. In fact, the comparativestructure is set already and much more information already is given with respectto T and 0 idioms. The child had only to fill in an empty slot in the comparisonor to figure out the relevant property that is to be transferred to the object orperson compared and that is typified by the metaphorical vehicle; for example,be as dumb as a fish makes use of a property of fishes, that is, not speaking, forreferring to someone who is able to be silent and keep secrets; in be as two dropsof water one has to guess what two things by definition identical could share, thatis, the relevant property at issue and so on.

CONCLUSION

In this chapter, I examined some of the core notions that constitute the theoreticalfield of idiom representation and comprehension: (a) the notion of literal mean-ing and its relationship with compositionality and figurative language, (b) theconcept of motivation in idioms, and (c) the role of speakers' intuitions aboutidiom meaning and structure. I presented a new set of linguistic and psychologi-cal studies on idioms that progressively abandon an "axiomatic idea," the non-compositionality of idioms. In order to clarify the possible roles of both thestructure of the idiom and the meanings of the words composing it on thestrategies people follow in interpretation, I presented some evidence comingfrom two studies: one concerned with the interpretative strategies adopted byadults when faced with unfamiliar idioms and the other with the interpretative


paths followed by children when asked to perform some metalinguistic tasks onidioms.

What is the picture that emerges from the literature and evidence I have beenexamining? At the very beginning, I mentioned that researchers in the fieldperceive a need to reform the traditional idea that idioms are no more than lexicalunits stored in the lexicon with no semantic as well syntactic structure. Themotivation for this change arises from experimental evidence as well as from theprogresses made in the area of meaning representation and processing. First,there is a need to differentiate between types of idioms according to their seman-tic as well syntactic characteristics. One relevant dimension certainly can be thedegree of semantic transparency of idioms, as reflected by both speakers' intui-tions and on-line processing measures. Second, we need to recognize that themeanings of the words forming an idiom and its underlying structure (being moreor less deeply rooted in the cognitive structure) are both available and play a rolewhen necessary for communicative purposes, that is, when one has, for instance,to understand unfamiliar idioms, or to produce a pragmatically motivated varia-tion on a classical idiom, and so forth. As a sort of processing default, it seemsstill plausible to speak in terms of activation of the meanings of the words and ofthe semantic description associated with an idiom. Still, the senses of the wordsare available, at least up to and even some time after an idiom's recognitionoccurs (cf. Cacciari & Tabossi, 1988; Tabossi & Cacciari, 1988), and these wordsenses are usable in cases such as the previous ones.

Just as an example, let us take an Italian idiom to make a hole in the water thatmeans to perform an activity in a rather unsuccessful way. Suppose one hears thefollowing sentence: "Doing X is like making holes in the water." In order toinfer the semantic domain to which the idiom's meaning makes reference (if notthe meaning in itself) and that serves as a basis for the intended judgment on theaction, one has to retrieve the meanings of the words composing it (lexical-locallevel), form a candidate interpretation of the figurative action (there is no plausi-ble literal meaning in this case), then reason that making holes in the water isquite an unsuccessful activity, in that it serves as a prototypical instance ofuselessness and infer, by analogy, the intended values of "doing X" (conceptual-analogical level). This example briefly illustrates some of the abilities and infor-mative sources that are at work when an unknown idiom is met. In everydayconversation, of course, we do not need to use such an amount of processing andinferential energies; nonetheless this is what understanding idioms is all about.

ACKNOWLEDGMENTS

I am grateful to The Council for International Exchange of Scholars for a Full-bright travel grant in 1987 that allowed me to think more deeply about idiom, toenjoy the hospitality of the Department of Psychology of Princeton University,

50 CACCIARI

and to start a wonderful collaboration with Sam Glucksberg, whom I againthank. Part of the results of the studies contained in this chapter was presentedfirst at the Syilvia Beach Conference on Language Comprehension in 1991. 1want to thank Morton Ann Gernsbacher for her hospitality and useful comments.

REFERENCES

Ackerman, B. P. (1982). On comprehending idioms: Do children get the picture? Journal of Experi-mental Child Psychology, 33, 439-454

Broderick, J. P. (1975). Metaphors in idioms: A problem for linguistic theory. The USF LanguageQuarterly, XIII, 3-4, 7-12.

Cacciari, C., & Glucksberg, S. (1990). Understanding idiomatic expressions: The contribution ofword meanings. In G. B. Simpson (Ed.), Understanding word and sentence (pp. 217-240).Amsterdam, Netherlands: Elsevier.

Cacciari, C., & Levorato, M. C. (1989). HoAs children understand idioms in discourse. Journal ofChild Language, 16, 387-405.

Cacciari, C., & Levorato, M. C. (1991, Novenmber). Spilling the beans on children's comprehen-sion and production of idioms. Paper presented at the 32nd Annual Meeting of The PsychonomicSociety, San Francisco.


Chafe, W. (1970). Meaning and the structure of language. Chicago: University of Chicago Press.Clark, H. H. (1978). Inferring what is mean. In J. M. Levelt & G. B. Flores d'Arcais (Eds.),

Studies in the perception of language (pp. 295-322). Chichester, England & New York: Wiley.Clark, H. H. (1979). Responding to indirect speech acts. Cognitive Psychology, 11, 430-477.Clark, H. H. (1983). Making sense of nonce sense. In G. B. Flores d'Arcais & R. J. Jarvella (Eds.),

The process of language understanding (pp. 297-331). Chichester, England: Wiley.Clark, H. H., & Carlson, T. B. (1981). Conmext for comprehension. In J. Long & A. Baddeley

(Eds.), Attention and performance (Vol. S), pp. 313-330). Hillsdale, NJ: Lawrence ErlbaumAssociates.

Clark, H. H., & Clark. E. (1977). Psychology and language. New York: Harcourt Brace Janovich.Clark, H. H., & Gerrig, R. J. (1983). Understanding old words with new meanings. Journal of

Verbal Learning and Verbal Behavior, 22, 591-608.Coulmas, F. (1981). Idiomaticity as a problem of pragmatics. In H. Parret & M. Sbisa' (Eds.),

Possibilities and limitations of pragmatics 'pp. 139-151). Amsterdam, Netherlands: John Ben-jamins.

Cronk, B. (1990, June). Familiarity and likedinessforfigurative and literal meanings of idioms.Poster presented at the 2nd Annual Convention of the American Psychological Association,Dallas.

Cruse, D. A. (1986). Lexical semantics. New York: Cambridge University Press.Dascal, M. (1987). Defending literal meaning. Cognitive Science, 11, 259-281.Davies, M. (1982/1983). Idiom and metaphor Proceedings of the Aristotelian Society, 83, 67-85.Fernando, C. (1978). Towards a definition of idiom, its nature and function. Studies in Language,

2(3), 313-343.Fernando, C., & Flavell, R. H. (1981). On idiom. Critical views and perspectives. Exter Linguistics

Studies, 5, pp. 1-S4.Fillmore, C. J. (1978). On the organization of the semantic information in the lexicon. In D. Farkas,

W. M. Jacobsen, & K. W. Todrys (Eds.), Papersfrom the parasession on the lexicon (pp. 148-173). Chicago: Chicago Linguistic Society.


Fillmore, C. J. (1979). Innocence: A second idealization for linguistics. Proceedings of the annualmeeting of the Berkeley Linguistic Society, (pp. 63-76).

Fillmore, C. J., Kay, P., & O'Connor, M. C. (1988). Regularity and idiomaticity in grammaticalconstructions: The case of "let alone." Language, 64, 501-538.

Fraser, B. (1970). Idioms within a transformational grammar. Foundations of Language, 6(1), 22-42.

Fromkin, V. A. (1971). The non anomalous nature of anomalous utterances. Language, 47, 27-52.Gaatone, D. (1981). Les locutions verbales: Pourquoi faire? Revue Romane, 16, 49-73.Gerrig, R. J. (1989). The time course of sense creation. Memory & Cognition, 17, 194-207.Gerrig, R. J., & Murphy, G. L. (1991, November). Contextual influences on the comprehension of

complex concepts. Poster presented at the 32nd Annual Meeting of The Psychonomic Society,San Francisco.

Gibbs, R. W. (1984). Literal meaning and psychological theory. Cognitive Psychology, 8, 191-219.

Gibbs, R. W., & Gonzales, G. P. (1985). Syntactic frozenness in processing and rememberingidioms. Cognition, 20, 243-259.

Gibbs, R. W., & Nayak, N. P. (1991). Why idioms mean what they do. Journal of ExperimentalPsychology: General, 120, 93-95.

Gibbs, R. W., Nayak, N. P., Bolton, J. L., & Keppel, M. (1988). Speakers' assumptions about thelexical flexibility of idioms. Memory & Cognition, 17(1), 58-68.

Gibbs, R. W., Nayak, N. P., & Cutting, N. (1989). Psycholinguistic studies on the syntacticbehavior of idioms. Cognitive Psychology, 21, 100-138.

Gibbs, R. W., & O'Brien, J. E. (1990). Idioms and mental imagery: The metaphorical motivationfor idiomatic meaning. Cognition, 36, 35-68.

Glucksberg, S. (1991). Beyond literal meanings. The psychology of allusion. Psychological Sci-ence, 2, 146-152.

Glucksberg, S., Gildea, P., & Bookin, M. B. (1982). On understanding non-literal speech: Canpeople ignore metaphors? Journal of Verbal Learning and Verbal Behavior, 21, 85-98.

Glucksberg, S., McGlone, M., & Cacciari, C. (in press). Semantic productivity and idioms: Dis-course processes.

Hobbs, J. (1979). Metaphor, metaphor schemata and selective inferencing. Technical Note No. 204SRI International, Menlo Park, CA.

Hockett, C. F. (1958). Idiom formation. In M. Halle et al. (Eds.), For Roman Jakobson (pp. 222-229). The Hague, Netherlands: Mouton.

Hoffman, R. R., & Honeck, R. P. (1980). A peacock looks at its legs: Cognitive science andfigurative language. In R. P. Honeck & R. R. Hoffman (Eds.), Cognition andfigurative language(pp. 3-24). Hillsdale, NJ: Lawrence Erlbaum Associates.

Holland, D., & Quinn, N. (Eds.). (1987). Cultural models in language and thought. New York:Cambridge University Press.

Honeck, R. P., & Hoffman, R. R. (1980). (Eds.). Cognition andfigurative language. Hillsdale, NJ:Lawrence Erlbaum Associates.

Katz, J. J. (1973). Compositionality, idiomaticity and lexical substitution. In S. R. Anderson & P.Kiparsky (Eds.), A festschrift for Morris Halle (pp. 357-376). New York: Holt, Reinhart &Winston.

Katz, J. J., & Fodor, J. (1963). The structure of a semantic theory. Language, 39, 170-210.Kreutz, R. J., & Graesser, A. C. (1991). Aspects of idiom interpretation: Comment on Nayak and

Gibbs. Journal of Experimental Psychology: General, 120(1), 90-92.Lakoff, G. (1986). The meanings of literal. Metaphor and Symbolic Activity, 1(4), 291-296.Lakoff, G. (1987). Women, fire and dangerous things. Chicago: University of Chicago Press.Lakoff, G., & Tirner, M. (1989). More than cool reason. A field guide to poetic metaphor. Chi-

cago: University of Chicago Press.

52 CACCIARI

Lehnert, W. G. (1986). The analysis of nominal compounds. VS, 44145, 155-180.Lehrer, A. (1974). Semanticfields and lexical structure. Amsterdam. Netherlands: North Holland.Levinson, S. C. (1983). Pragmatics. London: Cambridge University Press.Levorato, M. C., & Cacciari, C. (1992). Children's comprehension and production of idioms: The

role of context and familiarity. Journal of Child Language, 13, 415-433.Makkai, A. (1972). Idiom structure in English. The Hague, Netherlands & Paris: Mouton.Medin, D., & Shoben, E. (1988). Context and structure in conceptual combination. Cognitive

Psychology, 20, 158-190.Michiels, A. (1977). Idiomaticity in English. Revue des Langues Vivantes, 43, 184-199.Mueller, R. A. G., & Gibbs, R. W. (1987). Processing idioms with multiple meanings. Journal of

Psycholinguistic Research, 16(I), 63-81.Murphy, G. L. (1988). Comprehending complex concepts. Cognitive Science, 12, 529-562.Nagy, W. (1978). Some non-idiom larger-than-word units in the lexicon. In D. Farkas, W. M.

Jacobsen, & K. W. Todrys (Eds.), Papers fi-om the parasession on the lexicon (pp. 289-311).Chicago: Chicago Linguistic Society.

Nayak, N. P., & Gibbs, W. R. (1990). Conceptual knowledge in the interpretation of idioms.Journal of Experimental Psychology: Gen.ersl, 119, 115-130.

Nunberg, G. (1978). The pragmatics of rejerence. Bloomington: Indiana Linguistic Club.Nunberg, G. (1979). The non-uniqueness of semantic solutions: Polisemy. Linguistics and Philoso-

phy, 3(2), 143-184.Pollio, H. R., Bums, B. C. (1977). The anomaly of anomaly. Journal of Psycholinguistic Research,

6, 247-260.Popiel, S. J., & McRae, K. (1988). The figuraiive and literal uses of idioms, or all idioms are not

used equally. Journal of Psycholinguistic Research, 17(6), 475-487.Reagan, R. T. (1987). The syntax of English idioms: Can the dog be put on? Journal of Psycho-

linguistic Research, 16(5), 417-441.Rumelhart, D. E. (1979). Some problems with the notion of literal meaning. In A. Ortony (Ed.),

Metaphor and thought (pp. 78-90). New York: Cambridge University Press.Ruwet, N. (1983). Du bon usage des expressions idiomatiques. Recherches Linguistiques, 11, 5-

84.Schweigert, W. A. (1986). The comprehension of familiar and less familiar idioms. Journal of

Psycholinguistic Research, 15(1), 33-45.Searle, J. (1975). Indirect speech acts. In P. Cole & J. L. Morgan (Eds.), Syntax and semantics.

Speech acts (pp. 59-82). New York: Academic.Searle, J. (1979). Metaphor. in A. Orton) (Ed.), Metaphor and thought (pp. 92-123). New York:

Cambridge University Press.Swinney, D. A., & Cutler, A. (1973). The access and processing of idiomatic expressions. Journal

of Verbal Learning and Verbal Behavior, 18, 523-534.Tabossi, P., & Cacciari, C. (1988). Context effects in the comprehension of idioms. In Cognitive

Science Society (Ed.), Proceedings oj the lenth Annual Conference of the Cognitive ScienceSociety (pp. 90-96). Hillsdale, NJ: Lawrence Erlbaum Associates.

Tannen, D. (1989). Talking voices. Repetition, dialogue and imagery in conversational discourse.Cambridge, England: Cambridge University Press.

Ullmann, S. (1962). Semantics: An introducion to the study of meaning. Oxford, England: BasilBlackwell & Mott.

Wasow, T., Sag, I., & Nunberg, G. (1983). Idioms: An interim report. In S. Hattori & K. Inoue(Eds.), Proceedings of the XI11th International Congress of Linguistics (pp. 102- 105). Tokyo:CIPL.

Winner, E. (1988). The point of words. Children's understanding of metaphor and irony. Cam-bridge, England: Cambridge University Press.

APPENDIX A

List of experimental idioms:

1. Back to the salt mines.

2. Bow and scrape.3. Be dry behind the ears.

4. The fat is in the fire.5. Go to bed with the chickens.6. Have kittens.7. Out of the hole.8. Rap on one's knuckles.

9. Stick in one's throat.10. Take the stump.11. Throw oneself at someone's head.

12. Walk the chalk line.13. Be behind the eight-ball.14. Pay cash on the barrelhead.15. Explode a bombshell.

16. Feel one's oats.

17. Grease the palm.18. Lick one's boot.19. Have one's ears pinned back.20. Rest on one's oars.21. Straw in the wind.22. Take to the woods.23. Throw one's hat in the ring.

24. Where the shoe pinches.25. Be asleep at the switch.

26. The bottom drops out.27. Darken one's door.28. Fall down on the job.

29. Give her the gun.

30. Grease the wheels.31. Make the feathers fly.32. Polish the apple.33. Show a clean pair of heels.

53

54 CACCIARI

34. Take the bit in one's mouth.35. Tell it to the marines!36. Trip the light fantastic.

37. At one's heels.38. Hose someone down.

39. Down the line.40. Fan the breeze.41. Go by the board.

42. Hard sledding.43. The shoe is on the other foot.44. Pour oil on troubled waters.

45. Stare in the face.46. Take the starch out of someone.47. Throw cold water on someone.

48. Be under the hammer.

APPENDIX B

List of experimental idioms used in their original Italian form, with their word-by-word translation and meaning:Quasi-metaphorical idioms:

1. Avere paura anche delta propria ombra (be afraid of one's own shadow:"be afraid by everything").

2. Essere come due goccie d'acqua (be as two drops of water: "be verysimilar").

3. Essere noioso come una mosca (be as tedious as a fly: "be tedious").

4. Essere come cane e gatto (be like dog and cat: "be enemies").5. Essere muto come un pesce (be as dumb as a fish: "be silent and keep

secrets").6. Essere in un mare di guai (be in a sea of troubles: "be in many trou-

bles").7. Dormire come un ghiro (sleep as a dormouse: "sleep very soundly").

8. Sentirsi come un animale in gabbia (feel as an animal in a cage: "feelconstrained, limited in one's freedom").

9. Andare indietro come i gamberi (go backward as crayfishes: "do notprogress").


10. Sentirsi come un pesce fuor d'acqua (feel as a fish outside the water:"feel uneasy").

Transparent idioms:

1. Costare un occhio delta testa (cost an eye of the head: "cost a lot ofmoney").

2. Piangere sul latte versato (cry over spilled milk).

3. Fare un buco nell'acqua (make a hole in the water: "be unsuccessful").4. Cercare un ago in un pagliaio (look for a needle in a haystack).

5. Essere sulle nuvole (be among the clouds: "be absent-minded").6. Affogare in un bicchier d'acqua (drown oneself in a glass of water: "get

lost over nothing").7. Fare ridere i polli (make chickens laugh: "be socially unsuccessful").

8. Chiudere bottega (close shop).

9. Non sapere che pesci prendere (not know which fish to catch: "be veryuncertain").

10. Essere sulle spine (be on the thorns: "be very anxious").

Opaque idioms:

1. Darsi delta arie (give oneself some airs: "beast").2. Prendere per il naso (take someone by the nose: "tease someone").3. Essere al settimo cielo (be at the seven sky: "be in seventh heaven").

4. Avere del sale in zucca (have some salt in the pumpkin: "be intelligent").5. Rompere il ghiaccio (break the ice).

6. Toccare ii cielo con un dito (touch the sky with afinger: "be extremelyhappy").

7. Mangiare la foglia (eat the leaf: "understand a secret").8. Essere al verde (be at the green: "be broke").

9. Cadere dalle nuvole (fell from the clouds: "be astounded").10. Tagliare la corda (cut the rope: "escape").

3 Why Idioms Are NotDead Metaphors

Raymond W. Gibbs, Jr.University of California, Santa Cruz

Idioms suffer terrible indignities within linguistics, philosophy, and psychology.Compared to metaphors, which are thought to be "alive" and creative, idiomstraditionally have been viewed as dead metaphors or expressions that were oncemetaphorical, but that have lost their metaphoricity over time. Scholars generallyhave assumed that idioms exist as frozen, semantic units within speakers' mentallexicons in the same way that long, perhaps ambiguous, words are representedmentally. Unlike comprehension of literal language, idioms presumably are un-derstood through the retrieval of their stipulated meanings from the lexicon aftertheir literal meanings have been rejected as inappropriate (Bobrow & Bell, 1973;Weinreich, 1969), or in parallel to processing of their literal meanings (Estill &Kemper, 1982; Swinney & Cutler, 1979), or directly without any analysis of theirliteral meanings (Gibbs, 1980, 1985, 1986).

My purpose in this chapter is to show that the dead metaphor view of idi-omaticity is dead wrong. I argue that many idioms are very much alive meta-phorically, and that speakers make sense of idioms because of the metaphoricalknowledge that motivates these phrases' figurative meanings. This alternativeconception of idiomaticity is important not only for our understanding of figur-ative language use and comprehension, but more significantly, because it illus-trates the figurative nature of everyday thought. In the next section of this article,I discuss some of the reasons why scholars mistakenly assume that all idioms aredead metaphors. I then go on to describe why idioms are analyzable, and then Idiscuss empirical work that motivates why idioms mean what they do. Followingthis I show how metaphorical concepts influence people's understanding of idi-oms in different situational contexts. The concluding section presents mythoughts on the prospects for a psycholinguistic theory of idiom comprehension.

57

DEAD AND CONVENTIONAL METAPHORS

Scholars often treat idioms as dead metaphors because they confuse dead meta-phors with conventional ones. For example, suppose we encounter a word likegone in an expression like He's almost gone to speak of a dying person. Thetraditional dead metaphor theory would claim that gone is not really metaphoricnow, although it once may have been. Gone simply has come to have "dead" asone of its literal meanings. In a similar manner, the phrases spill the beans andkick the bucket are not viewed as metaphorical, though they at one time mighthave been quite figurative. Spill the beans simply has come to mean "reveal thesecret," whereas kick the bucket now simply means "to die." Each of theseidiomatic meanings presumably is listed as one of these phrase's literal meaningsalongside their other literal meanings that are based on their compositional analy-ses, such as "tip over the beans" and "to strike your foot against the pail"(Green, 1989).

Because contemporary speakers have little understanding of the original meta-phorical roots of phrases, such as spill the beans and kick the bucket, people arethought to comprehend idioms in the same way they know the meanings ofindividual words, as a matter of convention (cf. Lewis, 1969). For example, it isjust an arbitrary fact of the language or a convention that we use the term chair torefer to chairs because we easily could have used any other word. This samearbitrariness of meaning is seen with idioms. Thus, it is conventional within ourculture to greet someone by inquiring after the other person's health (e.g., Howare you? or How have you been?), whereas in some other cultures it is conven-tional to greet someone by asking about the other person's gastronomic welfare(e.g., Have you eaten?) (Morgan, 1978). It is to some extent arbitrary whether ornot a particular culture uses one form of greeting as opposed to another.

The meanings of idioms might be determined by such arbitrary conventions ofusage. Thus, the expression break a leg to wish a performer good luck before aperformance originated with the old superstition that it would be bad luck to wishsomeone good luck. Consequently, people started wishing their fellow actorsgood luck by wishing them bad luck (e. g., a broken leg). Over time the choice ofbreak a leg has become rigidly fixed as a convention, one reason why similarphrases will not serve the same purpose (e.g., fracture a tibia, I hope you breakyour leg). Contemporary speakers may now understand that break a leg means"to wish someone luck" simply as a matter of convention without any awarenessof why this phrase means what it does. It is for this reason that many idioms,such as break a leg, are considered to be dead metaphors.

Idioms are thought to have been metaphorical once because often we can traceback a phrase to its fully metaphorical uses in earlier stages of the language.Consider, for example, the phrasefork in the road. The wordfork in this phrasehas an established meaning just as it does in the phrase knife andfork. But we canwell imagine that at an earlier time when the word regularly was applied only to

58

3. WHY IDIOMS ARE NOT DEAD METAPHORS

the eating and cooking implement, people would use the word metaphorically inspeaking of a place in which a road divided into two parts, much like the tines ofan eating fork divide from its base into two or more separate directions. Peopledo not necessarily make use of this earlier metaphorical mapping between eatingimplements and divided paths when understanding fork in the road, again sug-gesting that we normally interpret idioms as dead metaphors (Alston, 1964).

One problem with this analysis offork in the road, as well as with many of ourintuitions about the historical development of idioms, is that it happens to bewrong (Kronfeld, 1980). As it turns out, the original meaning of fork (from theOld English forca and Latin furca) is not the eating or cooking implement. Thisis a later sense, first recorded in the 15th century. The original sense seems to bethe agricultural implement (pitchfork), and this developed in Middle English intoa more general sense: anything that forks, bifurcates, or divides into branches.Thus, fork in the road is not really a dead metaphor because it did not develop asa metaphoric extension of eating fork or of pitchfork for that matter. Rather, bothfork in the road and kitchen fork are specific tokens of the general sense "any-thing that forks," which is a concept that is very much alive in our conceptualsystems.

The misinterpretation about the development of idiomatic meaning illustrateshow difficult it is for speakers using language in ordinary circumstances or evenprofessionally reflecting about it to have valid intuitions about metaphoricity indiachronic processes. The closer we feel that a particular metaphor is dead, theharder it is to reconstruct the process through which the original metaphor wasunderstood. Our intuitions may not be sufficient to determine that some idiomswere once metaphorically alive, but now exist in a petrified state in the language.

Even words that appear to be classic examples of dead metaphors have vitallyalive metaphorical roots. Sweetser (1990) demonstrated that in Indo-Europeanlanguages, words meaning "see" regularly acquire the meaning "know" atwidely scattered times and places. The dead metaphor view of idiomaticityprovides no reason for why the same kinds of meaning change recur over andover again throughout the history of Indo-European languages. But one easilycan explain such changes in terms of conceptual metaphors (Sweetser, 1990). Inthe case of see words, there is a widespread and ancient conceptual metaphor thatKNOWING IS SEEING, which is part of the more general MIND-AS-BODYmetaphor (Sweetser, 1990). Because the metaphor exists in the conceptual sys-tems of Indo-European speakers, the conceptual mapping between seeing andknowing defines what Sweetser called a "pathway" for semantic change, so thatas new words for seeing develop, they eventually extend their meanings toknowing. The KNOWING IS SEEING metaphor, along with most other concep-tual metaphors, actually motivates why the language changes in regular waysthat make sense to us as speakers.

Determining whether an idiom is dead or just unconsciously conventionalrequires, among other things, a search for its systematic manifestation in the

59

60 GIBBS

language as a whole and in our everyday reasoning patterns (cf. Lakoff &Johnson, 1980). There are plenty of basic conventional metaphors that are alive,certainly enough to show that what is conventional and fixed need not be dead(Lakoff & Turner, 1989). Consider some of the many conventionalized ex-pressions about love. In English, speakers talk of love in the following ways: Hewas burning with love, I am crazy about her, We are one, I was given newstrength by her love, The magic is gone, Don't ever let me go, She pursued himrelentlessly, and so on. One might be tempted to argue that each phrase repre-sents some specific metaphor that is rio longer part of our everyday thinking.Many scholars assume that what makes each of these expressions seem fairlyliteral is that each one no longer possesses any degree of metaphoricity. But thereis a great deal of systematicity in these various expressions that reflects particularmetaphorical ways that we often think about love. For instance, I was given newstrength by her love, I thrive on love, He's sustained by love, and I'm starvedforyour affection reflect the metaphorical concept of love as some kind of nutrient.The LOVE AS NUTRIENT conceptual metaphor has as its primary function thecognitive role of understanding one concept (e.g., love) in terms of another(e.g., nutrients). Conceptual metaphors arise when we try to understand difficult,complex, abstract, or less delineated concepts, such as love, in terms of familiarideas, such as different kinds of nutrienl:s.

Theorists have not come to terms with the fact that much of ordinary languageis metaphoric because they hold the belief that all metaphors that are conven-tional and seemingly literal must be dead and really not metaphors any longer.This position fails to distinguish between conventional metaphors, which are partof our live conceptual system (e.g., LDVE IS A NUTRIENT), and historicalmetaphors that have long since died out. The mistake derives from an assumptionthat things in our cognition that are most alive and most active are those that areconscious. On the contrary, those that are most alive and most deeply en-trenched, efficient, and powerful are those that are so automatic as to be uncon-scious and effortless (Lakoff & Johnson. 1980). Our understanding of love as anutrient is active and widespread, but effortless and unconscious. Part of theevidence that conventional metaphors exist as live aspects of cognition is theiroccurrence in novel, poetic creations. The conceptual metaphor LOVE IS ANUTRIENT is beautifully elaborated upon in a poem titled "I drank a liquornever brewed" by Emily Dickinson:

I drank a liquor never brewedFrom tankards scooped in pearl.Not all the Frankfort berriesYield such an alcohol.

Inebriate of air am IAnd debauchee of dew,


Reeling through endless summer daysFrom inns of molten blue.

When landlords turn the drunken beeOut of the foxglove's door,When butterflies renounce their drams,I shall but drink the more,

Till seraphs swing their snowy hatsAnd saints to windows runTo see a little tipplerFrom the manzanilla come!

These lines instantiate in spectacular ways some of the rich set of entailmentsthat arise from the metaphorical mapping of nutrients onto love. If conceptualmetaphors, such as LOVE IS A NUTRIENT, did not exist in our conceptualsystems, then we could not understand novel, poetic language that makes use ofthem or even less creative phrases such as I'm drunk with love. In a similarfashion, our understanding that many idiomatic phrases have the figurativemeanings they do is due to our ability to conceptualize experience in a meta-phorical manner. The fact that idiomatic expressions, such as spill the beans,blow your stack, pop the question, and so on have highly conventionalizedmeanings does not in any way imply that these phrases are metaphorically dead.

IDIOMS ARE NOT NONCOMPOSITIONAL

The argument that the figurative meanings of idioms are not dead but meta-phorically conventional suggests that people somehow have insight into themeanings and lexical makeup of these phrases. This idea seems contrary to anyview that idioms are dead metaphors or that idioms are noncompositional be-cause their individual word meanings do not contribute to the overall figurativemeanings of these phrases (e.g., the words in kick the bucket do not appear tocontribute to this phrase's figurative meaning of "to die"). The noncomposi-tional nature of idioms suggests why idioms cannot undergo various syntacticoperations without loss of their nonliteral meanings. For example, John kickedthe bucket cannot be passivized into The bucket was kicked by John and stillretain the meaning that John died (cf. Gibbs & Gonzales, 1985). Furthermore,the dead metaphor view of idioms also seems consistent with the observation thatidioms appear to be lexically frozen. Thus, John kicked the bucket cannot bealtered lexically into John kicked the pail without disrupting its figurative mean-ing. In general, the frequent use of idioms with their noncompositional, figur-ative meanings makes their nonliteral interpretations highly conventional andlexicalized (Heringer, 1976), one reason why idioms appear to be understood

61

62 GIBBS

directly without analysis of their literal meanings (Gibbs, 1980, 1985, 1986;Ortony, Schallert, Reynolds, & Antos, 1978). Because the link between a idiomand its figurative meaning is conventionalized in the sense of being arbitrary,most speakers should have little ability to recognize why idioms have the mean-ings they do.

Part of the problem with most scholarly accounts of idiomaticity is that theylimit their discussion to only a small number of idiomatic phrases. There isplenty of discussion of classic phrases such as kick the bucket, but less consider-ation has been given to the wide range of expressions that are generally idi-omatic. The failure to consider the complete range of idioms results in in-complete theoretical generalizations about the nature of idiomaticity. Recentresearch, for instance, has shown that the individual words in many idiomssystematically contribute to the overall figurative interpretations of thesephrases, contrary to the noncompositional view of idioms (Fillmore, Kay, &O'Connor, 1988; Gibbs & Nayak, 1989; Lakoff, 1987; Langacker, 1986; Nun-berg, 1978). For example, the results of linguistic analyses and various experi-mental work in psycholinguistics have shown that American speakers know thatspill the beans is analyzable because beans refers to an idea or secret and spillingrefers to the act of revealing the secret. Similarly, in the phrase pop the question,it is easy to discern that the noun question refers to a marriage proposal when theverb pop is used to refer to the act of uttering it. Idioms such as pop the question,spill the beans, and lay down the law are "decomposable" because each of theircomponents obviously contributes to their overall figurative interpretations.Other idioms whose individual parts do not contribute individually to the figur-ative meaning of the idiom are semantically "nondecomposable" (e.g., kick thebucket, shoot the breeze) because people experience difficulty in breaking thesephrases into their component parts (Gibbs & Nayak, 1989; Nunberg, 1978). Athird group of idioms was viewed as decomposable, but abnormally so becausetheir individual components have a different relationship to their idiomatic refer-ents than do "normally" decomposable idioms. For example, we can identify thefigurative referent in the idiom carry a torch only by virtue of our knowledge oftorches as conventional metaphors for descriptions of warm feelings. Similarly,we understand the hitting of certain buttons in hit the panic button as a conven-tional metaphor for how we act in extreme circumstances. Each of these abnor-mally decomposable idioms differs from normally decomposable idioms, such asbutton your lips, whose components have a more direct relation to their figurativereferents.

The analyzability of an idiom does not depend on that word string beingliterally well formed (Gibbs & Nayak, 1989). For instance, pop the question isliterally anomalous but semantically decomposable. All that matters for an idiomto be viewed as decomposable is for its parts to have meanings, either literal orfigurative, that contribute independently to the phrase's overall figurative in-terpretation. The analyzability of an idiom is really a matter of degree depending


on the salience of its individual parts. Many idiomatic expressions exhibit inter-mediate degrees of analyzability. For instance, many speakers view the phraseplay with fire as being less decomposable than pop the question because themeaning that play contributes to play with fire is not as salient as the meaningthat pop contributes to pop the question. The more salient some word is within anidiom, the more likely that the word shares the same semantic field as does itsidiomatic referent. Thus, when speakers judge that the idiom let off steam isanalyzable or decomposable, they essentially are finding some relationship be-tween the components let off and steam with their figurative referents "release"and "anger." This relationship between an idiom's words and their figurativereferents is not arbitrary, but is based on active metaphorical mappings betweendifferent source and target domains that structure much of our experience. Forexample, the noun steam and the concept of anger are part of the same semanticfield because anger is understood metaphorically, in part, in terms of heat andinternal pressure (Gibbs & O'Brien, 1990; Lakoff, 1987; Lakoff & Johnson,1980). This metaphorical motivation for idioms is discussed further later.

People's intuitions about the analyzability of idioms play an important role indetermining these phrases' syntactic productivity (Gibbs & Nayak, 1989), lexicalflexibility (Gibbs, Nayak, Bolton, & Keppel, 1989), ease of comprehension(Gibbs, Nayak, & Cutting, 1989), and ease of learning (Gibbs, 1987, 1991). Forexample, most linguistic discussion on the syntactic deficiencies of idioms pro-vides little motivation as to why some idioms are productive and others not.Syntactically productive idioms are those that retain their figurative meaningswhen seen in a variety of syntactic constructions, whereas unproductive idiomsare "frozen" in that they lose their figurative interpretations when syntacticallyaltered.

The hypothesis examined in several recent studies was that people's intuitionsabout the syntactic versatility of idioms are affected by the analyzability ordecomposability of these figurative phrases (Gibbs & Nayak, 1989). Resultsfrom a variety of experiments demonstrated that normally decomposable idioms(e.g., pop the question) were found to be much more syntactically productivethan semantically nondecomposable idioms (e.g., chew the fat). Abnormallydecomposable idioms were not found to be syntactically productive because eachpart does not by itself refer to some component of the idiomatic referent, but onlyto some metaphorical relation between the individual part and the referent. Thus,readers rarely accept a passivized construction such as A torch for Sally wascarried by Jim as having a recognizable idiomatic meaning. These findingssuggest that the syntactic versatility of idioms is not an arbitrary phenomenon,perhaps due to unknown historical reasons (Cutler, 1982), but can be explained atleast partially in terms of an idiom's semantic analyzability. In a similar manner,the lexical flexibility of some idioms can be explained in terms of semanticanalyzability (Gibbs, Nayak, Bolton, & Keppel, 1989). Thus, the decomposablephrase pop the question was found to better retain its figurative meaning when

63

64 GIBBS

changed lexically (e.g., into either burst the question or pop the request) than didthe nondecomposable phrase kick the bucket when lexically altered (e.g., intoboot the bucket or kick the pail).

The analyzability of idioms also plays an important role in their immediate,"on-line" interpretations. Because the individual components in decomposableidioms contribute systematically to the figurative meanings of these phrases,people may process idioms in a compositional manner where the semantic repre-sentations of each component are accessed and combined according to the syn-tactical rules of the language. For example, the phrases lay down and the lawserve as cues to the retrieval of the figurative meaning for lay down the lawbecause each of these components has an independent meaning contributing tothe entire idiom's nonliteral interpretation. On the other hand, a strict composi-tional analysis of semantically nondecomposable idioms (e.g., kick the bucket,chew the fat, go for broke) provides little information about the figurative mean-ings of these expressions. Consequently, readers or listeners subsequently mighthave to recover the directly stipulated meanings of nondecomposable phrasesfrom the mental lexicon, a process that requires additional processing effort.

A series of reading-time studies pro-vided some support for these hypotheses(Gibbs, Nayak, & Cutting, 1989). The results of these experiments showed thatpeople took significantly less time to process decomposable idioms than to readthe nondecomposable expressions. Both normally and abnormally decomposablephrases took less time to process than their respective literal, control phrases(e.g., pop the question versus ask the question), but nondecomposable idiomsactually took longer to process than their respective literal controls (e.g., chewthe fat versus cook the fat). These data suggest that people attempt to do somecompositional analysis when understanding idiomatic phrases. When an idiom isdecomposable readers can assign independent meanings to its individual partsand quickly will recognize how these meaningful parts combine to form theoverall figurative interpretation of the phrase. These processing differences in thecomprehension of decomposable and nondecomposable idioms do not imply thatreaders have no directly stipulated figurative meanings for decomposable idioms.Instead, it appears that the analyzability of decomposable idioms provides a veryuseful source of information that facilitates people's recognition that an idiomaticword string is meant to have a figurative interpretation. One explanation for thecommonly observed finding that idioms are processed more quickly than literalphrases is that these studies primarily employ idioms that are more analyzablethan nondecomposable (cf. Gibbs, Nayak, & Cutting, 1989).

One of the most debated issues i~n the research on idiom comprehensionconcerns the role that literal meaning plays in people's understanding of thesefigurative phrases. Previous research has demonstrated that readers do not needto analyze first the literal meaning of an entire idiom phrase before determiningits nonliteral interpretation (Gibbs, 1980, 1985, 1986; Gibbs & Gonzales, 1985;Ortony et al., 1978; Swinney & Cutler, 1979). However, people may process the


literal meanings of the individual words simultaneously to determining the over-all figurative interpretation of these phrases (Cacciari & Tabossi, 1988). Onemight assume that the compositional analysis that people perform on idioms isbased on the literal meanings of their individual words. Decomposable idiomsmight be easier to comprehend than nondecomposable phrases because the literalmeanings of decomposable expressions directly contribute to these phrases' fig-urative meanings. On the other hand, processing the literal meanings of seman-tically nondecomposable idioms should be uninformative and might interferewith people's comprehension of these phrases.

This possibility about the role of literal meaning in idiom comprehensionraises two significant issues. First, there is the problem of specifying the literalmeanings of the individual words in idioms. Consider the idiomatic phrase overmy dead body. What are the literal meanings of the individual words in thisexpression? The preposition over has over 100 related senses that are motivatedby different metaphoric and metonymic principles (Brugman & Lakoff, 1989). Itis unclear which of these meanings is literal, nor is it clear which of thesemeanings actually is activated when the phrase over my dead body has beenunderstood. Most psycholinguistic models that assume that the literal meaningsof words are activated during idiom processing fail to specify exactly what thesemeanings are or how such meanings are used in determining the figurativeinterpretations of idioms. Without some specification of what literal meaning isin general, or what constitutes the literal meanings of individual words, it seemspremature to posit that people automatically process idioms by analyzing theliteral meanings of its individual words. Again, I argue that people recognize thatmost idioms are at least partially compositional because their individual compo-nents have independent meanings. But this does not imply that these individualword meanings are literal or context-free. Instead, people at least appear torecognize that the individual parts of idioms have some figurative meanings thatcontribute to the overall nonliteral interpretations of idiomatic phrases.

A second issue regarding literal meaning and idiom processing concerns thefact that many idioms are literally ill-formed or opaque, such as by and large, inthe know, crack a joke, and promise the moon. Literally ill-formed idiomsviolate selectional restriction rules. For instance, the idiom swallow his pride isliterally ill-formed or anomalous because it violates the selection restriction thatthe verb swallow only occurs in sentences with noun phrases referring to somephysical object. Most linguistic analyses assume that expressions that do notpossess well-defined literal meanings must receive their interpretations by stip-ulation in the lexicon (Dowty, Wall, & Peters, 1982). But many ill-formedidioms are normally decomposable (e.g., pop the question, perish the thought),whereas many semantically nondecomposable idioms are literally well formed(e.g., chew thefat, hit the sauce, give the sack). If people perform a composi-tional analysis on an idiom based on its entire literal meaning, then people shouldprocess literally well-formed idioms faster than they do ill-formed expressions.

65

66 GIBBS

However, the data from Gibbs, Navak, and Cutting (1989) clearly demon-strate that this is not the case. Participants in these studies actually took less timeto understand ill-formed idioms than they did literally well-formed expressions.It appears that understanding idioms only requires that people assign figurativemeanings to the parts of idioms; there is no need to analyze automatically eachexpression according to its entire literal interpretation. This seems especiallylikely given people's extreme familiarity with many idiomatic expressions. Thuspeople ordinarily attempt to perform some sort of compositional analysis, al-though not necessarily a literal analysis, when comprehending idiom phrases toattach meanings to these phrases' specific parts (Gibbs, Nayak, & Cutting,1989). It is important to restate this point. The figurative meanings of idiomsmay be based on their internal compositional semantics even though this does notmean that idiomatic meaning is based on what scholars normally assume is literalmeaning. Contrary to the popular conception that the literal meaning of a phraseor sentence is its compositional meaning, many phrases have compositionalmeanings that are based on the figurative meanings of their individual parts.

WHY IDIOMS MEAN WHAT THEY DO

There is substantial experimental evidence, then, that the meanings of idioms canbe motivated partially in that speakers recognize some, often figurative, rela-tionship between the words in idioms and their overall figurative interpretations.Earlier I suggested that the words in analyzable idioms share the same semanticfield as do their figurative referents. The parts of idioms refer to different knowl-edge domains, many of which are conceptualized in terms of metaphor. Peoplemay recognize tacitly that the metaphorical mapping of information between twoconceptual domains actually motivates why idioms mean what they do. Forexample, the idiom John spilled the beans maps our knowledge of someonetipping over a container of beans to a person revealing some previously hiddensecret. English speakers understand spill the beans to mean "reveal the secret"because there are underlying conceptual metaphors, such as the MIND IS ACONTAINER and IDEAS ARE PHYSICAL ENTITIES, that structure theirconceptions of minds, secrets, and disclosure (Lakoff & Johnson, 1980). Eventhough the existence of these conceptual metaphors does not predict that certainidioms or conventional expressions must appear in the language, the presence ofthese independent conceptual metaphors by which we make sense of experienceprovides a partial motivation for why specific phrases (e.g., spill the beans) areused to refer to particular events (e.g., the revealing of secrets).

Linguistic analyses of idioms provide some grounds for believing that idiomsdo not exist as separate semantic units within the lexicon, but actually reflectcoherent systems of metaphorical concepts. For example, the idiomatic phrasesblow your stack, flip your lid, hit the ceiling, get hot under the collar, lose your


cool, and get steamed up appear to be motivated by the conceptual metaphorANGER IS HEATED FLUID IN A CONTAINER, which is one of the small setof conceptual mappings between different source and target domain that formpart of our conceptualization for anger. But is there any evidence that metaphorssuch as ANGER IS HEATED FLUID IN A CONTAINER are really conceptualand not, more simply, generalizations of linguistic meaning? Much of my workin the past few years has been devoted to exploring the possibility that idiomsmake sense in the ways that they do precisely because they are motivated byconceptual knowledge that is metaphorical. Thus, idioms like blow your stackandflip your lid make sense to us as referring to the idea of getting very angry ina way that phrases like run to the store or mow the lawn would not. The deadmetaphor view suggests no reason for why some idioms seem so appropriate inhaving the meanings they do. After all, if the meaning of a phrase is truly dead,then people should not believe that any particular ways of talking about some-thing like anger, such as hit the ceiling or blow your stack, seem so reasonable ina way that mow your lawn or go to the store do not.

One way to uncover speakers' tacit knowledge of the metaphorical basis foridioms is through a detailed examination of speakers' mental images for idioms(Gibbs & O'Brien, 1990; Lakoff, 1987). Consider the idiom spill the beans. Tryto form a mental image for this phrase and then ask yourself the followingquestions: Where are the beans before they are spilled? How big is the container?Are the beans cooked or uncooked? Is the spilling accidental or intentional?Where are the beans once they have been spilled? Are the beans in a nice, neatpile? Where are the beans supposed to be? After the beans are spilled, are theyeasy to retrieve?

Most people have definite responses to these questions about their mentalimages for idioms. They generally say that the beans were in some pot that isabout the size of a person's head, the beans are uncooked, the spilling of thebeans is accidental, the spilled beans are all over a floor and are difficult toretrieve. This consistency in people's intuitions about their mental images isquite puzzling if one assumes that the meanings of idioms are determined ar-bitrarily. Gibbs and O'Brien (1990) used people's descriptions about their mentalimages for idioms as a way of discovering some of the metaphorical knowledgethat motivates the meanings of idiomatic phrases. We examined people's mentalimages for groups of idioms with similar figurative meanings, such as aboutrevelation (e.g., spill the beans, let the cat out of the bag, blow the lid off), anger(e.g., blow your stack, hit the ceiling, flip your lid), insanity (e.g., go off yourrocker, lose your marbles, bounce off the walls), secretiveness (e.g., keep itunder your hat, button your lips, keep in the dark), and exerting control (e.g.,crack the whip, lay down the law, call the shots). Participants were asked todescribe their mental images for these idioms and to answer questions about thecauses, intentionality, and manner of actions in their mental images for thesephrases.

67

68 GIBBS

We expected a high degree of consistency in participants' descriptions of theirmental images for idioms with similar meanings because of the constraints con-ceptual metaphors (e.g., the MIND IS A CONTAINER, IDEAS ARE PHYS-ICAL ENTITIES, and ANGER IS HEAT l impose on the links between idiomaticphrases and their nonliteral meanings. If people's tacit knowledge of idioms isnot structured by different conceptual metaphors, there should be little consisten-cy in participants' responses to questions about the causes and consequences ofactions within their mental images of idioms with similar nonliteral interpreta-tions.

People had little difficulty with the mental imagery task. Most of the report'sparticipants gave for their mental images of idioms contained rich details. Forinstance, one individual reported that her mental image for call the shots was asfollows: "An army sergeant standing in front of a line of soldiers all of whom arefaced in the same direction and when the sergeant shouts very loudly the soldiersall commence firing their rifles." Overall, participants' descriptions of theirmental images were remarkably consistent for different idioms with similar figur-ative meanings. Across the five groups of idioms we studied, 75% of partici-pants' mental images for the different idioms described similar general images.These general schemas for people's images were not simply representative of theidioms' figurative meanings, but captured more specific aspects of the kinestheticevents with the images. For example, the anger idioms such as flip your lid andhit the ceiling all refer to the concept of "getting angry," but participants specifi-cally imagined for these phrases some force causing a container to release pres-sure in a violent manner. There is nothing in the surface forms of these differentidioms to tightly constrain the images participants reported. After all, lids can beflipped and ceilings can be hit in a wide variety of ways, caused by manydifferent circumstances. But our participants' protocols revealed little variation inthe general events that took place in their images for idioms with similar mean-ings.

Participants' responses to the questions about the causes and consequences ofthe actions described in their images were also highly consistent (over 88% whenaveraged across both the different probe questions and the five groups of idioms).Consider the most frequent responses to the probe questions for the anger idioms(e.g., blow your stack, flip your lid, hit the ceiling). When imagining angeridioms, people reported that pressure (i.e., stress or frustration) causes the ac-tion, that one has little control over the pressure once it builds, its violent releaseis done unintentionally (e.g., the blowing of the stack), and that once the releasehas taken place (i.e., once the ceiling has been hit, the lid flipped, the stackblown), it is difficult to reverse the action.

Why are people so consistent in their intuitions about the causes and conse-quences of the actions described in their mental images for idioms? We arguedthat people are limited in the kinds of images they create for idioms because of


very specific conceptual knowledge that is mostly metaphorical. For example,the participants' images for the anger idioms are based on folk conceptions ofcertain physical events. For anger, people use their knowledge about the behaviorof heated fluid or vapor building up and escaping from containers (ones that ourparticipants most frequently reported to be the size of a person's head). Thus,people's metaphorical mapping of knowledge from a source domain (e.g., heatedfluid in a container) onto target domains (e.g., the anger emotion) helps themconceptualize in more concrete terms what is understood about the target domainof anger. Various specific entailments result from these general metaphoricalmappings, ones that provide specific insight into the causes, intentionality,manner, and consequences of the activities described by stacks blowing, lidsflipping, and ceilings being hit. Our argument is that the metaphorical ways inwhich we partially conceptualize experiences, such as anger, actually providepart of the motivation for why people have consistent mental images, andspecific knowledge about these images, for idioms with similar figurative mean-ings.

Of course, any consistency in people's mental images for idioms with similarfigurative meanings might be due not to the constraining influence of conceptualmetaphors, but, more simply, to the very fact that these phrases share nearidentical meanings. Thus, people are constrained in their mental images for blowyour stack, flip your lid, hit the ceiling, and so on because they all mean "to getvery angry," not because they are motivated by similar conceptual metaphors.The figurative meaning "to get angry" by itself does not convey much informa-tion about the causes and consequences of the actions described in people'smental images so we did not feel this alternative hypothesis carried much weight.A set of follow-up studies showed that knowing the figurative meaning of anidiom (e.g., "getting angry") does not by itself account for why people havesuch systematic knowledge of their images of idioms (e.g., blow your stack orflip your lid). Asking people to imagine "to get very angry" produces, notsurprisingly, a wide variety of mental images. Most importantly, though, ourparticipants showed very little consistency in their responses to the questionsregarding causation, intentionality, manner, and consequences about the actionsin their mental images for paraphrases of the idioms. Furthermore, people weremuch less consistent in their mental images for literal phrases (e.g., blow yourtire) than for idioms (e.g., blow your stack) because they do not possess the samedegree of conceptual knowledge about their images for literal phrases as they dofor idiomatic expressions.

The results of these mental imagery studies support the idea that the figurativemeanings of idioms are motivated by various conceptual metaphors that existindependently as part of our conceptual system. Traditional theories of idi-omaticity have no way of accounting for these imagery findings because theyassume that the meanings of idioms arise from metaphors that are now dead.

69

CONCEPTUAL METAPHOR AND IDIOMINTERPRETATION

Most of our concepts referring to ideas about anger, love, insanity, and revelationare not structured by a single conceptual metaphor, but are organized partially bya variety of metaphorical mappings. For example, our conceptualization of loveis structured by various conceptual metaphors, such as LOVE IS A JOURNEY,LOVE IS A NUTRIENT, and LOVE IS WAR, each of which gives rise todifferent entailments about our love experiences (Kovecses, 1986). Similarly,anger is not just understood in terms of a heated fluid in a container, but also canbe conceptualized in terms of ANGER IS ANIMAL BEHAVIOR and ANGER ISINSANITY. The mere existence of these independent conceptual mappings doesnot predict necessarily that certain idioms or conventional expression shouldexist as part of the language. But we can easily find examples of idioms thatappear to be motivated by different conceptual metaphors. For instance, theidioms blow your stack and bite your head off both express extreme degrees ofanger. However, bite your head off makes sense because people can link thelexical items in this phrase to the conceptual metaphor ANGRY BEHAVIOR ISANIMAL BEHAVIOR. An animal jumping down a victim's throat is similar tosomeone shouting angrily. On the other hand, people understand the figurativemeaning of blow your stack through the conceptual metaphor ANGER ISHEATED FLUID IN A CONTAINER where a person shouting angrily has thesame explosive effect as does the top of a container blowing open under pres-sure.

Psycholinguistic research has shown that people's knowledge of the meta-phorical links between different source and target domains provides the basis forthe appropriate use and interpretation of idioms in particular discourse situations(Nayak & Gibbs, 1990). Participants in one study were asked to read stories thatreflected different conceptual metaphors about a particular emotion concept. Onepair of stories is presented here, each of which conceptualizes anger in differentways. The first is as follows:

Mary was very tense about this evening's dinner party.The fact that Bob had not come home to help was making herfume. She was getting hotter with every passing minute.Dinner would not be ready before the guests arrived.As it got close to five o'clock the pressure was reallybuilding up. Mary's tolerance was reaching its limits.When Bob strolled at ten minutes to five whistling andsmiling, Mary

blew her stackbit his head off

70


This story was written to prime the metaphorical mapping ANGER IS HEAT-ED FLUID IN A CONTAINER by depicting Mary's increasing anger in terms ofincreasing pressure and heat. The use of phrases such as very tense, making herfume, getting hotter, the pressure was really building up, and reaching its limitsare specific references to this mapping. Another story primed a different concep-tual metaphor, ANGER IS ANIMAL BEHAVIOR:

Mary was getting very grouchy about this evening's dinner party.She prowled around the house waiting for Bob to come home to help.She was growling under her breath about Bob's lateness.Her mood was becoming more savage with every passing minute.As it got closer to five o'clock Mary was ferociously angry with Bob.When Bob strolled in at 4:30 whistling and smiling, Mary

bit his head offblew her stack

Participants in this study were asked to read each story and rate the appropri-ateness of each final phrase given its preceding story context. The results indi-cated that readers' judgments about the appropriateness of an idiom in context areinfluenced by the coherence between the metaphorical information depicted in adiscourse situation and the conceptual metaphor reflected in the lexical makeupof an idiom. Thus, participants gave higher appropriateness ratings to blew herstack in a story that described the woman's anger as being like heated fluid in acontainer whereas bit his head off was seen as more appropriate in a story thatdescribed the woman's anger in terms of a ferocious animal. Even though twoidioms may have highly similar figurative meanings, this does not imply thattheir meanings are identical because of the different conceptual metaphors thatmotivate their figurative interpretations. The phrase bite his head off impliesthat the angry person is demonstrating her anger in a more deliberate, intentionalmanner than is the case with the phrase blow her stack. This conclusion aboutdifferences in the meanings of similar idioms is significant because it providessome motivation for why some idioms seem especially appropriate to use insome situations but not others. The traditional dead metaphor view provides noexplanation for the context-sensitivity of idioms other than to suggest that the useof idioms is an arbitrary matter of convention. The conceptual view of idi-omaticity suggested here offers a motivated reason for why idioms mean whatthey do and are used in specific kinds of discourse situations.

Another consequence of the conceptual view of idiomaticity is that idiomsshould not be identical in meaning to their literal paraphrases. Contrary to thedead metaphor view, which assumes that the meanings of idioms are arbitrary forspeakers and can be represented mostly in short phrases or even single words(Palmer, 1981), idioms have rather complex interpretations. For example,

71

72 GIBBS

phrases such as spill the beans can not be paraphrased simply as meaning "toreveal a secret" in the way that most idioms dictionaries do (cf. Boatner, Gates,& Makkai, 1975; Long & Sumrhers, 1979). The mapping of source domains suchas containers onto minds results in very specific entailments about the act ofrevealing a secret. Thus, the act of revealing a secret usually is seen as beingcaused by some internal pressure within the mind of the revealer, the action isthought to be done unintentionally, and the action is judged as being performed ina forceful manner (Gibbs & O'Brien, 1990). One interesting possibility is thatpeople actually draw these inferences about the act of revealing a secret eachtime they comprehend the idiom phrase spill the beans. However, people mightbe less likely to draw such inferences about causation, intentionality, and mannerwhen comprehending literal paraphrases of idioms, such as reveal the secret.Literal phrases, such as reveal the secret, are not motivated by the same set ofconceptual metaphors as are specific idioms such as spill the beans. For thisreason, people do not view the meanings of spill the beans and reveal the secretas equivalent despite their apparent similarity.

I have recently tested this hypothesis about the nonequivalence of idioms andtheir literal paraphrases in a series of experiments (Gibbs, 1990, 1992). Partici-pants read stories that described different human events, such as revealing se-crets, getting angry, losing control of themselves, and so on. These stores con-tained information about the causes of the event, the intentionality of the actionperformed by each story's protagonist, and the manner in which the actions wereperformed. Some stories depicted this information in a manner that was consis-tent with the entailments of particular conceptual metaphors (the no-violationcontexts). Presented next is an example of a no-violation story:

John heard some interesting gossip about Paul and Mary.Even though Paul and Mary were married to other people,they had recently started having a passionate affair.John was very surprised when he found out about the affair.So John called up another friend who knew Paul and Maryand accidentally blurted out what he knew.The friend commented to John that he had really

spilled the beans.(or)

revealed the secret.

Note that each of the entailments about the cause, intentionality, and mannerin which a secret is revealed is stated explicitly and correctly in this story. Each ofthese entailments arises from the conceptual metaphors the MIND IS A CON-TAINER and IDEAS ARE PHYSICAL ENTITIES (Gibbs & O'Brien, 1990).


Now consider a story from one of the violation conditions in which one of theoriginal entailments (Intentionality) has been altered:

John heard some interesting gossip about Paul and Mary.Even though Paul and Mary were married to other people,they had recently started having a passionate affair.John was very surprised when he found out about the affair.John fully intended never to say a word to anyone.One day he was talking to someone who knew Paul and Mary,when John purposefully said something about what he knew.The friend commented to John that he had really

spilled the beans.(or)

revealed the secret.

The empirical question was whether people understand idioms and their literalparaphrases in different ways depending on whether a story context is consistentor inconsistent with the specific entailments of the conceptual metaphors thatmotivate the meanings of those idioms. In the first study, participants simplyrated the appropriateness of each final phrase given the preceding story. A secondexperiment measured the speed with which idioms and their paraphrases wereprocessed in different contexts.

The results showed that the participants rated the idioms (e.g., spill the beans)and their literal paraphrases (e.g., reveal the secret) as being equally appropriatein the no-violation story contexts. However, the participants judged idioms asbeing less appropriate in the different violation conditions than they did literalparaphrases in these same contexts. Literal paraphrases are not constrained byconceptual metaphors in the way idioms are limited. Consequently, phrases suchas reveal the secret were seen as appropriate in most story contexts regardless ofthe cause of the revelation, the intentionality of the act, or the manner in which itis done. This was not the case for spill the beans. Data from the reading-timeexperiment showed that idioms took longer to process when they were read at theend of violation contexts than in the no-violation stories, whereas the literalparaphrases were roughly as easy to process in the violation stories as in the no-violation conditions.

These findings demonstrate that idioms are not equivalent in meaning to theirsimple literal paraphrases. Idiomatic phrases have very specific figurative mean-ings that result from the entailments of the underlying conceptual metaphors thatmotivate their figurative interpretations. Literal phrases, such as reveal the se-cret, are not motivated by the same conceptual metaphors and consequently areless specific in meaning. In general, these data provide further evidence againstthe idea that idioms are dead metaphors and equivalent in meaning to theirputative, literal paraphrases.

73

CONCLUSION

The dead metaphor view has dominated the study of idiomaticity in linguistics,philosophy, and psycholinguistics. By assuming that idioms are simple deadmetaphors, figurative language researchers have had reason to neglect idioms infavor of other tropes, such as metaphor, that appear to be more creative or alive.But research in the cognitive sciences over the past few years clearly has demon-strated the fatal difficulties with the dead metaphor view of idiomaticity. Manyidioms are not simple dead metaphors that are noncompositional in meaning. Norare the figurative meanings of many idioms arbitrarily stipulated and listed assimple, literal paraphrases in the mental lexicon. Instead, idioms are partiallycompositional and their rich figurative meanings are motivated by the meta-phorical knowledge people possess of the domains to which idioms refer. Thesemetaphorical mappings between source and target domain knowledge often areconventionalized in the sense that they are so much a part of our everydaycognition as to be unconscious and automatic. Such figurative mappings providepart of the link between the lexical makeup of idioms and their figurative mean-ings such that many idioms make sense in having the meanings they do. Further-more, the context-sensitive use and interpretation of idioms, at least, is dueprimarily to the recognition that many idioms are motivated by different meta-phorical mappings that exist independently in our conceptual systems.

The conceptual view of idioms presented here offers a reasonable alternativeto the traditional, dead metaphor view of idiomaticity. It is important to realize,however, that the dead metaphor and conceptual views of idiomaticity should notbe seen as competing theories. Many idiomatic phrases very well could be deador have meanings that are determined arbitrarily as matters of convention. At thesame time, there exist a wide range of idioms that are extragrammatical, such asby and large, all of a sudden, and take advantage of, that resist a conceptualanalysis. My argument, though, is that many, perhaps thousands of idiomsprofitably can be understood as being motivated partially by figurative schemesof thought, ones that are very much alive and part of our everyday thinking andreasoning. The scholarly study of idioms no longer can afford to view thesephrases as a homogeneous class of linguistic items that differ from literal lan-guage, that differ from metaphor, and that resist standard grammatical analysis.Idiomatic language is remarkably complex and each phrase demands its ownanalysis in terms of its syntactic, semantic, pragmatic, and conceptual proper-ties.

Previous experimental work on idiom comprehension in psycholinguistics hastended to emphasize issues such as whether literal meanings contribute to idiomprocessing, whether context facilitates recognition of idiomatic meanings, andwhether the frequency or familiarity of idioms plays significant roles in howthese phrases are learned and understood. The theoretical models that have beenexamined in these earlier empirical studies generally assume that there is, or

74


should be, a single answer to the question of how idioms are understood. Ibelieve that this assumption is seriously wrong. As we learn more about thecomplexity of idiomatic phrases, it seems increasingly likely that no singletheory or model can account for all kinds of idioms and all kinds of discoursesituations (both conversational and literary). Instead, different sources of lin-guistic and nonlinguistic information are evaluated during processing that con-tribute to people's immediate understanding of what speakers mean when theyuse idiomatic phrases. This idea suggests, for example, that listeners do notalways instantiate specific conceptual metaphors that motivate an idiom's mean-ing when understanding some phrase in conversation. Similarly, people may notalways analyze the literal word meanings of idioms during comprehension.There will be occasions when people do tap into an idiom's conceptual founda-tion. Readers also might process the individual word meanings when they at-tempt to comprehend certain kinds of idioms. But it is a mistake to assume thatsome types of analyses will occur each and every time someone encounters anidiomatic expression. The complexity of idioms leaves us, at present, with noother alternative than to embrace the plurality of theories that have been proposedto account for different aspects of idiomatic language.

The study of idioms is significant not only in terms of understanding howpeople learn and comprehend figurative language, but also because idiomaticityreveals some dramatic insights into the relationship of language and thought. Theprevailing view of mind holds that thought and language are inherently literal,and that both can be characterized in objective terms. My work on idiomaticityadvances the idea that the traditional view of mind is mistaken because humancognition is shaped fundamentally by various processes of figuration. It seemsclear now that idioms do not represent the graveyard of metaphorical thought.Instead, our use and understanding of idioms attests to the continuous process ofpoetic thinking.

ACKNOWLEDGMENTS

Preparation of this chapter was supported by Grant MH42980 from the NationalInstitute of Mental Health and by a Faculty Research Grant from the Universityof California, Santa Cruz.

REFERENCES

Alston, W. (1964). The philosophy of language. Englewood Cliffs, NJ: Prentice-Hall.Boatner, M., Gates, J., & Makkai, A. (1975). A dictionary of American idioms. New York: Baron's

Educational Series.Bobrow, S., & Bell, S. (1973). On catching on to idiomatic expressions. Memory & Cognition, 1,

343-346.

75

76 GIBBS

Brugman, C., & Lakoff, G. (1989). Cognitive topology and lexical networks. In G. Cottrell, S.Small, & M. K. Tanenhaus (Eds.), Lexical ambiguity resolution (pp. 477-508). Stanford, CA:Morgan.


Cutler, A. (1982). Idioms: The older the colder. Linguistic Inquiry, 13, 317-320.Dowty, D., Wall, R., & Peters, S. (1982). Introduction to Montague grammar. Dordrecht,

Netherlands: Reidel.Estill, R., & Kemper, S. (1982). Interpreting idioms. Journal of PsycholinguisticResearch, 9, 559-

568.Fillmore, C., Kay, P., & O'Connor, M. (1988) Regularity and idiomaticity in grammatical con-

structions: The case of let alone. Language, 64, 501-538.Gibbs, R. (1980). Spilling the beans on understanding and memory for idioms in conversation.

Memory & Cognition, 8, 449-456.Gibbs, R. (1985). On the process of understanding idioms. Journal of Psycholinguistic Research,

14, 465-472.Gibbs, R. (1986). Skating on thin ice: Literal meaning and understanding idioms in conversation.

Discourse Processes, 9, 17-30.Gibbs, R. (1987). Linguistic factors in children's understanding of idioms. Journal of Child Lan-

guage, 14, 569-586.Gibbs, R. (1990). Psycholinguistic studies on the conceptual basis of idiomaticity. Cognitive Lin-

guistics, 1, 417-451.Gibbs, R. (1991). Semantic analyzability in children's understanding of idioms. Journal of Speech

and Hearing Research, 34, 613-620.Gibbs, R. (1992). What do idioms really mean Journal of Memory and Language, 31, 385-406.Gibbs, R., & Gonzales, G. (1985). Syntactic frozenness in processing and remembering idioms.

Cognition, 20, 243-259.Gibbs, R., & Nayak, N. (1989). Psycholinguistic studies on the syntactic behavior of idioms.

Cognitive Psychology, 21, 100-138.Gibbs, R., Nayak, N., Bolton, J., & Keppel, M. (1989). Speakers' assumptions about the lexical

flexibility of idioms. Memory & Cognition, 17, 58-68.Gibbs, R., Nayak, N., & Cutting, C. (1989). How to kick the bucket and not decompose: Ana-

lyzability and idiom processing. Journal oj'Memory and Language, 28, 576-593.Gibbs, R., & O'Brien, J. (1990). Idioms and mental imagery: The metaphorical motivation for

idiomatic meaning. Cognition, 36, 35-68.Green, G. (1989). Pragmatics and natural language understanding. Hillsdale, NJ: Lawrence

Erlbaum Associates.Heringer, J. (1976). Idioms and lexicalization n English. In M. Shibtani (Ed.), Syntax or semantics:

The grammar of causative constructions (Vol. 6, pp. 205-216). New York: Academic Press.Kovecses, Z. (1986). Metaphors of anger, pride, and love. Amsterdam, Netherlands: John Ben-

jamins.Kronfeld, X. (1980). Novel and conventional metaphors: A matter of methodology. Poetics Today,

2, 13-24.Lakoff, G. (1987). Women, fire, and dangerous things. Chicago: Chicago University Press.Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago: Chicago University Press.Lakoff, G., & Turner, M. (1989). More than cool reason: A field guide to poetic metaphor. Chi-

cago: University of Chicago Press.Langacker, R. (1986). Foundations of cognitive grammar (Vol. 1). Stanford, CA: Stanford Univer-

sity Press.Lewis, D. (1969). Convention. Cambridge, MA: Harvard University Press.Long, T., & Summers, D. (1979). Longman dictionary of English idioms. London: Longman.

3. WHY IDIOMS ARE NOT DEAD METAPHORS 77

Morgan, J. L. (1978). Two types of convention in indirect speech acts. In P. Cole (Ed.), Syntax andsemantics: Vol. 9. Pragmatics (pp. 261-280). New York: Academic.

Nayak, N., & Gibbs, R. (1990). Conceptual knowledge in the interpretation of idioms. Journal ofExperimental Psychology: General, 119, 315-330.

Nunberg, G. (1978). The pragmatics of reference. Bloomington: Indiana University LinguisticsClub.

Ortony. A., Schallert, D., Reynolds, R., & Antos, S. (1978). Interpretating metaphors and idioms:Some effects of context on comprehension. Journal of Verbal Learning and Verbal Behavior, 17,465-477.

Palmer, F. (1981). Semantics, Cambridge, England: Cambridge University Press.Sweetser, E. (1990). Semantic structure and semantic change: A cognitive semantic linguistic study

of modality, perception, speech acts, and logical relations. Cambridge, MA: Cambridge Univer-sity Press.

Swinney, D., & Cutter, A. (t979). The access and processing of idiomatic expressions. Journal ofVerbal Learning and Verbal Behavior, 18, 523-534.

Weinreich, U. (1969). Problems in the analysis of idioms. In J. Puhvel (Ed.), Substance andstructure of language (pp. 23-81). Los Angeles: University of California Press.

4 The Comprehension andSemantic Interpretationof Idioms

Giovanni B. Flores d'ArcaisMax-Planck-lnstitute for Psycholinguistics, Nijmegen, TheNetherlands

This chapter deals with the comprehension of phrasal idioms. As is pointed out inseveral chapters in this book, idiomatic expressions represent a broad range oflinguistic phenomena, ranging from lexical idioms, such as rail at, to phrasalidioms such as to kick the bucket.

Although most idioms are phrases constructed according to syntactic well-formedness principles, a few contain syntactic violations, such as by and large.Throughout this chapter, I am concerned only with phrasal idioms. Thus, for thepresent purpose, an idiom can be defined as a phrasal unit whose meaning is notobtainable from its syntactic components on the basis of Frege's phrasal principleof compositionality.

An idiom can represent a simple or a complex lexical unit, which in severalcases can be taken to be synonymous with a single word (e.g., to pull someone'sleg means "to tease"), or can be expressed with a whole proposition (e.g., totake the bull by the horns means "to attack a problem without fear").

A phrasal idiom thus typically consists of a verbal phrase with a verb and anoun phrase (NP) (to kick the bucket). These expressions have acquired a particu-lar status, in the sense that the complete idiom has a given meaning that is notequal to the meaning of the parts as they are combined in the idiomatic ex-pression. In this sense, the "literal meaning" of the idiom, whenever available,usually has little or nothing to do with the idiomatic meaning. To attempt to makea syntactic decomposition of an idiom means generally to lose the idiomaticmeaning of it.

In this chapter I first summarize some of the properties of the idiomaticphrases, then briefly mention the main theories on the comprehension of idioms,and discuss some issues concerning idiom interpretation. The last part of the

79

80 FLORES D'ARCAIS

chapter reviews a number of experiments carried out as contributions toward atheory on the process of comprehending idiomatic phrases.

In the next sections, I first discuss a few important properties of idioms, whichin turn is useful for the ensuing discussion of the comprehension processes.

SOME IMPORTANT PROPERTIES OF IDIOMS

Frozenness and Flexibility

Idioms differ to the degree to which they can tolerate morphological and syntac-tic operations. Parts of some idioms can be quantified, modified, or even omit-ted, lexical elements can be inserted at various points, clauses can be embeddedin idiomatic phrases, and so forth. The degree to which this is possible dependson the degree of frozenness of an idiom. To say, for example, "the bucket waskicked by John" is syntactically perfectly possible, but the passivization opera-tion destroys the idiomatic meaning for the original expression. Certain mor-phological and syntactic operations are allowed for virtually all idioms (e.g.,insertion of an auxiliary in the phrase; cf. "he has kicked the bucket" vs. "hekicks the bucket"), whereas others are possible only with certain ones.

The extent to which an idiom can undergo syntactic and morphological opera-tions while at the same time retaining its idiomatic meaning can be expressed interms of the number and types of grammatical operations that an idiom is capableof tolerating. This degree can indicate the amount of frozenness of an idiom.Idioms can be classified on the basis of their degree of frozenness, from veryfrozen to very flexible. A kind of frozenness hierarchy has been proposed byFraser (1970). In this hierarchy, an idiom at a given level of such a hierarchy,capable of tolerating a given operation. also can tolerate all operations that areallowed by idioms "below" that level.

Transparency and Opacity

A second important distinction among idioms concerns the degree of their trans-parency. Idioms can be more or less opaque, or, conversely, more or less trans-parent. In transparent idioms, such as to leave the cat out of the bag, the literalmeaning is available, whereas in an opaque idiom such as to take a leak theliteral interpretation is no longer available or has never been or is not evenpossible. Most opaque idioms are fossils in which the literal meaning is no morethan a philological curiosity.

The two dimensions just considered, frozenness and flexibility on the onehand, and opacity and transparency on the other, represent continuous and nottrue dichotomies between two extreme and discrete values. An idiom can bemore or less frozen on the one hand, and more or less opaque on the other.Moreover, the two dimensions are likely to be not completely independent.

4. COMPREHENSION AND INTERPRETATION OF IDIOMS

Opaque idioms tend to be more frozen, although this tendency to covary seemsrather weak.

The Point of Idiom Uniqueness

One of the questions asked in the present chapter concerns the moment in whichan idiomatic phrase is recognized. In order to give an answer to this question, wehave to know at what point it is possible to recognize the idiom. At which pointdoes an idiom become an idiom?

Consider the following sentence with two different endings: In the discussionabout the financial situation, it became apparent to John that his arguments wereweak, and that he was going to lose {money/face}. It is clear that the phrase tolose face takes its idiomatic character only at the end of the sentence. It isunlikely that the recognition of the idiom exemplified here can take place beforethis point is reached in the sentence. In other cases, it is possible that an idi-omatic phrase be identified at the beginning or before its completion. For exam-ple, except in very specific contexts, the idiom to take the bull by the horns islikely to be identified once the phrase "the bull" has been read or heard.

We thus can introduce an important notion, namely of the point of idiomuniqueness, which is the point at which the idiom becomes uniquely identifiable.For every idiomatic phrase, there must be a point at which the phrase can beinterpreted only as an idiom. For many idiomatic phrases this point will be thelast word of the idiom.

Depending on the familiarity of the idiom and especially on the context, thepoint at which the reader/listener is able to correctly interpret the idiom may notcoincide with the point of idiom uniqueness, but in some cases can precede it.Consider, for example, the idiom to spill the beans (meaning "to violate asecret"). Without context, the point of idiom uniqueness is given by the wordbeans. However, with a previous context introducing the theme of the violationof a secret, the meaning of the idiom is likely to become available at the wordspill. I would like to call the point at which the meaning of the idiom is retrievedby the reader/listener as the point of idiom recognition, which, depending on thepresence of a context and on the familiarity of the idiom, can precede the point ofidiom uniqueness.

The point at which the idiomatic meaning is retrieved suggests a parallel to thepoint of sentence disambiguation. In fact, many idiomatic phrases can be consid-ered like ambiguous structures to which two alternative meaning interpretationscan be assigned. There is, however, a basic difference with situations of syntacticambiguity: While for syntactic ambiguities the two alternative interpretationsrequire two alternative parsing operations that produce two different grammaticalstructures, the two interpretations of an idiom, the literal and the idiomatic one(when both are possible), are normally consistent with the same syntactic analy-sis. In most cases, moreover, the two meanings, the one corresponding to the

81

82 FLORES D'ARCAIS

idiomatic and the one corresponding to the literal reading of the phrase, are notincompatible with each other-being related to each other by some form ofmetaphorical extension.

Syntactic Parsing and Idiomatic AnalysisAn idiomatic phrase normally is presented as part of a larger sentence structure.Whether each idiom is represented in the mental lexicon as a separate unit or not,an idiomatic phrase presents a syntactic structure on its own, and various mor-phological and syntactic operations are allowed in the idiomatic phrase. There isno reason to assume that, when an idiomatic phrase is encountered, normalsyntactic parsing normally is suspended or will not take place. Because thelinguistic processor cannot be sure about the nature of the idiom until at somepoint within the idiomatic phrase itself-normally until a point of "idiomuniqueness"-it is unlikely that the parser will be waiting for a whole "multi-word lexical entry" to be retrieved before resuming syntactic analysis.

The meaning of each sentence is obtained as a result of the operations of alexical processor and of a parser. The construction of the meaning of any sen-tence can be taken to be based on the principle of compositionality. The phrase"a big dog" is interpreted as some composite of the meaning of big and of themeaning of dog. But although the meaning of the phrase "to kick the ball" iscomposed on the basis of one of the meanings of kick and one of the meanings ofball, the meaning of to kick the bucket is not instantiated compositionally. Thus,we face a particular problem. A phrasal idiom is embedded in a sentence, andnormally the reader/listener does not know that the phrase ahead will be an idiomuntil a certain critical point-usually well into the idiomatic phrase-is reached.Thus, syntactic parsing should take place normally at least up to a given point inthe sentence. When a listener hears the sentence ". . . and then hekicked .... ", he or she does not yet know what the next NP is going to be, sonormal syntactic analysis should take place at least up to the point where the fullidiomatic expression is available, or when enough of it is available to allow anunique identification.

Thus, on the one hand an idiomatic phrase should be subject to normalsyntactic analysis, but on the other hand the result of parsing would be aninterpretation that is the literal one, based on the principle of compositionality,whereas the idiomatic character of the idiomatic phrase is not compositional, andthe meaning of the idiom is not instantiated on the basis of the joint operations ofthe parser and of the lexical processor.

The Lexical Status of Idioms and TheirRepresentation in the Mental LexiconWe have seen that the meaning of an idiomatic phrase is not to be reconstructedfrom the meaning of the elements that compose it. What, then, is the structure ofthis meaning? How are idioms represented in the mental lexicon? Three different

4. COMPREHENSION AND INTERPRETATION OF IDIOMS 83

answers can be given to these questions: (a) Idioms can be listed in the mentallexicon as multiword lexical entries; (b) idioms are not listed as such, but arereconstructed on the basis of the idiomatic meaning starting from the entry of oneor more of the content words that constitute the idiom; and (c) the meaning of theidiom would not be represented as such at all, but would be computed each timeon the basis of the lexical units and of the phrasal structure by some process ofmetaphor construction.

The first alternative (a) has taken the form of the idiom list hypothesis, whichwas proposed within linguistics by Weinreich (1966). According to this hypoth-esis, in addition to the "normal" dictionary entries the lexicon would contain alist of idioms. A normal dictionary entry would be made up of a set of phonologi-cal, syntactic, and morphonemic features, together with a sense description. Theidiom list would contain a string of morphemes of various length, together withthe meaning associated with it. In order to take this hypothesis seriously, we haveto assume double listing for words that are part of an idiom. These would belisted in the normal dictionary entry list and also in the idiomatic list. Of course,the idiomatic list hypothesis is also consistent with the idea that each of theentries of the mental lexicon that are part of an idiom would have a "pointer" tothe idiomatic meaning, much as in many dictionaries there is a special entryunder one of the main words of the idiom.

According to the second alternative (b), the idiom would not be listed asseparate entry, but would be reconstructed on the basis of one or more of itsindividual lexemes, which would be characterized by appropriate pointers to theidiomatic meaning.

The third alternative (c) is that idioms are not listed nor reconstructed, butcomputed every time they are met, via processes of analogical inference. Theprocess of idiom comprehension then would be equivalent to the process of thecomprehension of a metaphor.

Notice that it is by no means necessary that only one hypothesis holds. As wewill see, it is perfectly possible to expect differences in the representation and inthe availability of idiomatic phrases characterized by different levels of famil-iarity. Highly familiar idioms are likely to be represented as a unitary entry in themental lexicon, whereas for infrequent and unknown idioms, especially opaqueones, it is rather unlikely that such a unitary representation is available. Newidioms are likely to be treated by the listener/reader as a metaphor.

The three hypotheses have taken different forms in the literature (see, e.g.,Cacciari & Glucksberg, 1991, for a review; see also other chapters of this book),and are not discussed further in this chapter.

THE COMPREHENSION OF IDIOMATICEXPRESSIONS

Much as for other forms of figurative language, during the last 15 or 20 years anumber of psycholinguistic studies have been dedicated to the processes of

84 FLORES D'ARCAIS

comprehension of idiomatic expressions. Most psycholinguistic studies of idi-oms have been concerned with one, or both, of the following questions. The firstis whether the reader or listener retrieves both the literal and the idiomaticmeaning of the idioms, or only the idiomatic, whereas the second question askswhich of the two is computed first. After initial evidence in favor of the "literalfirst" hypothesis (Bobrow & Bell, 1973), or the "both meanings are computed inparallel" hypothesis (Swinney & Cutler, 1979), the bulk of recent evidence (see,e.g., Gibbs, 1984, 1986; Schweigert. 1986) seems to indicate that the idiomaticmeaning is accessed directly, at no extra costs, and that the literal meaning maynot even be recovered. On the other hand, these conclusions take diverse formsdepending on the degree of frozenness of the idiom and on the type of idiomstudied (see Gibbs, 1986). Some of this work is reported in other chapters of thisbook, so no attempt is made here to review it. Rather, in this chapter, I ignorethese two questions altogether, and instead look directly at some of the featuresthat are likely to characterize the process of idiom comprehension.

The Process of Idiom Comprehension

Before introducing some experimental work, I would like to propose a fewassumptions on the process of idiom comprehension, which have guided theempirical studies discussed in this chapter:

Assumption 1: The comprehension of an idiom does not require a specialprocessing mode. An assumption underlying much psycholinguistic work onidioms is that literal and idiomatic meaning are computed separately yielding twoalternative interpretations. This assumption may not hold. Most words and mostutterances are characterized by a whole range of senses or meanings, including"metaphoric" meanings. When talking about a person, the adjective warm usu-ally indicates a property of character or personality, and only in specific contextsit is used to indicate temperature of the body. Thus, the so-called idiomaticmeaning of a word within an idiom is likely to be obtained much as the othersenses of the word. All senses or meanings can be taken to become availableduring comprehension, and the "idiomatic" sense becomes available when theidiom is interpreted as such. This assumption relates to the following one.

Assumption 2: Lexical access is obligatory for individual words even withinidioms. As recent evidence has shown (e.g., Cacciari & Tabossi, 1988), indi-vidual lexemes always are accessed during language comprehension, even whenthey are part of an idiom. Thus individual words can facilitate the access ofwords related in meaning, independent of the fact that they are part of an idiomor not. The existence of priming effects of single words of an idiomatic phrase onwords related to their literal meaning (e.g., Tabossi & Cacciari, 1988) does notconstitute definitive evidence for the claim that a literal interpretation of the


idiomatic phrase is entertained during sentence comprehension, but certainlysupports the notion of an obligatory access, in the mental lexicon, to the singlelexemes that constitute the idiom.

Assumption 3: Syntactic analysis is obligatory. Recent work on human pars-ing has shown that the syntactic structure of any sentence is computed for anystring (see, e.g., Frazier, 1987). Syntactic computation is assumed to proceed inan automatic way (see Flores d'Arcais, 1982). Although a syntactic analysis isper se not necessary to understand the meaning of the idiom, the human parseralways would analyze it fully. This claim has been tested in one of the experi-ments reported in this chapter.

Let me now briefly sketch how I propose that idiomatic phrases be understoodwithin a sentence context. The underlying claim here is that the process oflanguage comprehension does not differ whether the sentences contain an idi-omatic phrase or not:

1. The syntactic parser analyzes the input strings and produces a gram-matical structure with a given representation.

2. The lexical processor accesses the lexical items in the mental lexicon anddelivers a meaning representation of the individual words. These two compo-nents can be taken to be modular, with the lexical processor feeding to the parserthe results of its analysis.

3. Semantic analysis is made on the basis of the syntactic structure obtainedand the appropriate lexical units associated with their meaning.

4. The results of the processes of these various components are taken andevaluated at the level of a message or thematic interpretation unit. Such a unithas been proposed as a feature characterizing several recent psycholinguisticmodels, with various terms used to identify it (e.g., thematic processor; Ferreira& Clifton, 1986; Frazier, 1987; Rayner, Carlson, & Frazier, 1983). Its function isessentially to collect, compare, and interpret evidence available from the lin-guistic components.

5. Even when the sentence contains an idiomatic phrase, according to thepresent view the syntactic parser and the lexical processor would work "nor-mally" throughout the process of analyzing the input, the parser yielding asyntactic structure upon which the appropriate lexical representations computedby the lexical processor would be inserted. At the point of idiom identification,the lexical processor should retrieve the unique lexical representation of theidiom and its corresponding meaning. Thus, some conflict could arise at thispoint, in principle but not necessarily, between the syntactic processor and thelexical processor. The interpretation available so far to the message or thematicinterpretation unit could become inconsistent with the new information sent bythe lexical processor. This inconsistency may result in some processing diffi-culty. It is likely that such difficulty is only present with nonfamiliar idioms.

85

86 FLORES D'ARCAIS

Highly familiar ones would become immediately available even at the lexicallevel and would not constitute any problem. Because most metaphors are under-stood without many difficulties, unfamiliar idioms also should present no prob-lems for understanding. On the other hand, an idiom could be computationallymore complex because of a possible conflict between the operation of differentprocessing components, the lexical processor, the syntactic parser, and the se-mantic analyzer.

The Problems

The points discussed in the previous section lead to a series of questions, towhich the present work has tried to offer some answers. A first question is tosome extent related to the issue frequently asked in psycholinguistic experiments,namely whether the literal meaning is computed during the comprehension pro-cess. Whether the literal meaning is computed or not, there could be a need ofadditional computational effort required to comprehend an idiomatic phrase ascompared to the comprehension of a nonidiomatic phrase. If the literal meaningis computed during idiom comprehension-be it before or in parallel with theidiomatic meaning-one should expect additional computational effort for com-prehending an idiom. But even if the literal meaning is not computed, an idiomcould be computationally more complex because of the possible conflict betweenthe operations of the different processing components.

A second question concerns a possible difference between well-known andpoorly known idioms. Highly familiar, frequently used items might be listed inthe lexicon, whereas the comprehension of idioms with low familiarity and thatare used rarely might require a full computation. An adequate psycholinguisticmodel of comprehension of idiomatic phrases should account for this possibledifference.

At the intersection of the two previous questions, a third one emerges. Theproblem whether the literal meaning is computed or not during idiom com-prehension is probably not appropriate without further specification. For veryfamiliar idioms it seems very unlikely that the literal interpretation is computed.For unfamiliar idioms, on the other hand, the story looks different. Highlyfamiliar idioms should not produce any processing problems. Low-familiarityidioms should present some difficulty, when the reader/listener is not yet in theconditions of finding out whether the phrase that she or he is reading is an idiom.Therefore, additional processing problems could arise at the moment a literalinterpretation becomes no longer possible, namely at the point of idiomuniqueness, and more strongly when the context is rather neutral. So, one canexpect some difficulty in processing low-familiarity idioms at the point of idiomuniqueness.

A fourth question, related to the assumption of an obligatory, automatic actionof the human parser, is whether full analysis of the grammatical structure of the


incoming string is taking place even when the sentence being understood in-cludes a highly familiar idiomatic phrase that could be recovered as a multiwordunit in the mental lexicon, without any need of syntactic analysis of its internalstructure. This hypothesis has been tested in an experiment by using a word-by-word reading task, reported later. The assumption based on the present hypoth-esis is that inspection times for words at critical points in the string should belonger, and that this should be the case for the point of idiom uniqueness ofunfamiliar idioms.

A fifth question, mentioned earlier in the present discussion, concerns thesemantic analysis of rarely used or completely unknown idioms. According to anumber of available theories on idioms in linguistics and psycholinguistics,idioms are listed in the lexicon, and are not computed. Whereas metaphoricexpressions would be comprehended by various inferential processes, idiomswould be searched in the lexicon as multiword lexical units. If this position istrue, then every time a new idiom is met, it should act as a completely unknownword, and the listener/reader would be requested to use the context to make upsome meaning for the phrase, or use some external source such as a dictionary ora human informant. However, because most idioms-except for extremelyopaque ones-bear some semantic or pragmatic relation to the intended mean-ing, listeners/readers should be in the condition of interpreting unknown idiomsand often obtain a plausible or not too inappropriate interpretation. This assump-tion has been tested in the last two studies reported in this chapter.

THE STUDIES

The present chapter reports on five different studies. A first study included anumber of rating and scaling experiments designed to obtain information aboutfamiliarity and about the point of idiom uniqueness for a large number of idi-omatic phrases. The second study tested the assumption that idiomatic phrasesthat, according to the list hypothesis, could be stored as multiword lexical units,and therefore could be interpreted as such, still undergo full syntactic analysis. Athird study tested the hypothesis that highly familiar idioms are recognized easilywithout any need for computing the literal meaning, whereas low-familiarityidioms require additional processing effort. A fourth and fifty study tried to findout how people interpret unknown idioms and which principles they may use inassigning such an interpretation.

Study 1: Ratings of Properties of Idiomatic Phrases

The purpose of this study was to obtain some indications of a number of proper-ties of a large sample of Dutch idioms. This study was carried out with a largenumber of subjects in different sections, and the details of the study are not

87

88 FLORES D'ARCAIS

reported here. The data obtained, besides being interesting in their own right,were used for the selection of the materials for the following studies reported inthe present chapter.

Two hundred Dutch idiomatic phrases were chosen as the basis material forour study. They were submitted in various sets for different tasks to a total of 294subjects, with the request to perform a number of tasks.

Definition of Meaning. The first task for the subjects was to give a definitionof the meaning of each of the idioms. The definitions given then were evaluatedby two judges on a 3-point scale as (1) correct, (2) partially correct or somewhatrelated to the "dictionary meaning" of the idioms, or (3) completely incorrect.The average scale values of the idioms computed over all subjects gave anindication of the availability or knowledge of the meaning of the idioms.

Familiarity. A second rating task consisted of giving an estimate of thefrequency with which the subject thought a given idiom would be used in thelanguage, by using a 7-point scale from "very frequently used" to "never used."This yielded for each idiom a "subjective frequency" or "familiarity" score.

Use of Literal Meaning. In a third task, subjects were requested to answerthe question whether in making an interpretation of the meaning of an idiom theythought that they had retrieved the "literal" meaning of the idiom. Following theanswer to this question, the subject had to express on a 5-point scale the degreeof confidence of having retrieved the literal meaning. The answers to thesequestions provided an indication of the extent to which the idiomatic meaning ofidioms might be accessed via a literal interpretation. The values based on thistask also were taken as a rough estimate of the degree of transparency of anidiom.

The Point of Idiom Uniqueness. In a fourth task, subjects were requested toindicate for each idiom the point, thar. is, the word, at which the idiom becomesuniquely defined.

Results. The definitions given and the ratings provided in the first place anumber of details about properties of the idioms, which were used as a basis forthe other studies. The full results of these ratings are reported elsewhere (Floresd'Arcais, in preparation). Some specific results are also of interest here.

On the basis of the definitions given in Task 1, the idioms were divided intotwo categories, namely the "well known" (high proportion of correct definition)and "not known" (incorrect definitions). Not surprisingly, "knowledge" and"familiarity" were highly correlated (r = .78). Correlations coefficients betweenthe average scale values thus obtained for the various idioms were computedseparately for the well-known idioms, and for the unknown ones. For the well-


known idioms, the correlation between knowledge and use of the literal in-terpretation was - .81, whereas for the unknown idioms the correlation betweenthe same variables was -. 30 (both coefficients are significantly different from0). Thus, when an idiom is well known, the reader does not think that he or sheuses the literal meaning in order to understand it, whereas when the idiom is notwell known, literal interpretation, according to the subjects' evaluation, is likelyto be used in many cases. Furthermore, for well-known idioms, the most familiarones are somewhat unlikely to be interpreted literally (correlation between famil-iarity scale values and use of literal meaning in interpretation r = -. 23, p <.05), whereas for unknown idioms familiarity and use of literal meaning do notcorrelate at all (r = .02).

Study 2: Syntactic Processing During Comprehensionof Idiomatic Phrases

If idiomatic phrases are stored in memory as multiword lexical units, as the listhypothesis claims, then syntactic analysis of the phrasal structure of the idiom inprinciple should not be necessary for comprehending the idiom. The presentexperiment tested the hypothesis that syntactic analysis is an obligatory, automat-ic process that does not depend on the particular structure being processed andthat is insensitive to the lexical properties of the phrasal structure being pro-cessed. If an idiom is recognized at an early point, further analysis of theidiomatic phrase is in principle no longer necessary to understand the meaning ofthe phrase. Thus, a full syntactic analysis of the idiomatic phrase would not berequired. If this is true, then a highly familiar idiom, which can be recognizedearly, could be understood without the need of a full syntactic analysis, whereas anonfamiliar idiom would be understood only after a full grammatical analysis.Thus, the experiment investigated the question of whether syntactic analysis ofthe idiomatic phrase continues even when the idiom has been recognized and theappropriate idiomatic meaning has been assigned to it.

To what extent is syntactic analysis performed during comprehension of thesentences containing idioms? An answer to this question was sought in an experi-ment, in which readers were requested to detect syntactic violations in sentencescontaining phrasal idioms of high or low familiarity.

Material. Twenty-four sentences containing idiomatic phrases (12 highlyfamiliar and 12 unfamiliar idioms) and 60 filler sentences constituted the materialof the present experiment. In 50% of the presentation each sentence contained agrammatical violation, such as the following:

1. Violation of gender agreement between article and noun. For example,instead of the appropriate idiom in hetzelfde schuitje, the phrase presentedincluded the masculine/feminine article dezelfde.

90 FLORES D'ARCAIS

2. Concordance singular plural: Twee vlieg in een klap instead of tweeviegen in een klap.

Subjects and Procedure. 20 students at the University of Leiden took part inthe experiment. The sentences were presented on a computer display one word ata time, in a left to right sequence with each word displayed occupying a differentposition as in a normal printed sequence. However, each word was presented andremained on the display only 250 ms and then disappeared. The end of thesentence was signalled by a period that could be accompanied by an asterisk tosignal to the subjects to make a decision whether the sentence was correct orcontained some violations.

Results. The results, proportions of successfully detected violations in high-and low-familiarity idiomatic phrases, are summarized in Table 4.1. In bothcases, there is the same probability of hitting a violation. The latencies for thecorrect responses were significantly different for familiar and unfamiliar idioms:Correct reports of grammatical violations took significantly shorter with highlyfamiliar than with unfamiliar idioms [F(1, 23) = 4.96, p < .051.

Discussion. The absence of any difference in the rate of detection of syntac-tic violations in high-familiarity and low-familiarity idioms can be interpreted byassuming that in both cases syntactic analysis is taking place. We had expected alower proportion of detection of violations after the point of idiom recognitionfor high-familiarity items, under the assumption that once the idiom has beenrecognized, the reader no longer would pay attention to the surface structure ofthe rest of the phrase. Instead, syntactic violations still are detected, thus allow-ing the conclusion that syntactic analysis of the phrasal string that constitutes theidiom goes on even when the high-familiarity item has been recognized. Thisresult is consistent with work by Cacciari and Tabossi (1988), which also has

TABLE 4.1Proportion of Error Detection and Average Response Latencies (ms)

for Correct Detection of Syntactic Violations in High-and Low-Familiarity Idioms

Idiomatic Phrases

High Low SignificanceFamiliarity Familiarity Difference

Proportion Errors .28 .31 Not significantAverage Latencies 1278 1102 p < .05

(Correct Responses)


shown that syntactic analysis is carried out automatically, whenever an inputstring is given for comprehension.

Study 3: Word-By-Word Reading of SentencesContaining Idiomatic Phrases

As we already have discussed, if comprehending idiomatic phrases requiresadditional processing, because of the need of computing first a literal and then anidiomatic representation, or the necessity of constructing two interpretations inparallel, as several of the models in the literature alternatively suggest, then weshould expect increasing processing loads with idiomatic phrases as comparedwith nonidiomatic ones. According to our previous discussion, however, we donot expect this to be the case for familiar idioms, for which no extra processingload should be expected. On the other hand, low-familiarity idioms might pro-duce some processing problems when the reader/listener is at the point in whichthese can be uniquely identified as idioms. This should be more the case for theidioms embedded in a neutral context. Thus, in an appropriate experiment wepredict some processing difficulty for the low-familiarity idiomatic phrases at thepoint of idiom uniqueness, especially when a neutral context does not offerprevious cues for the appropriate idiomatic interpretation of the phrase. Further-more, opaque, nonfamiliar idioms should produce more processing difficultythan transparent ones.

Thus the present experiment tested the hypothesis that highly familiar idiomsdo not produce any processing difficulty, whereas unfamiliar idioms should yieldsome processing difficulty at the points of idioms uniqueness. As an indication ofprocessing difficulty we took the time of inspection of each word in a word-by-word reading task.

On the basis of the norms obtained in the first study summarized previously, Ichose a number of highly familiar and unfamiliar idioms. For each of the twocategories I selected a number of transparent and a number of opaque idioms. Foreach of the experimental idiomatic phrases, I had available an indication of thepoint of idiomatic uniqueness, obtained on the basis of the ratings of a number ofsubjects, as previously reported. These words were taken as the critical points forthe measurement of reading latencies in the present experiment. This materialwas used in a self-paced reading experiment, in which sentences containing theidioms were presented one word at a time on a display, under subject's control.The intervals between the pressing responses of the subject were measured andrecorded.

Material. Twenty-four idiomatic phrases, 12 familiar and 12 very unfamiliarones, of which respectively 6 were transparent and 6 very opaque, constituted theexperimental material. They were embedded in neutral, literal or idiomatic con-

91

92 FLORES D'ARCAIS

text sentences. Seventy-two filler sentences were intermixed with these 24 sen-tences containing idioms. Twenty-four of these fillers were matched sentenceswith identical syntactic structures and very similar contents to the sentencescontaining idioms. The sentences were presented on a display one word at thetime, under subjects' control.

Subjects and Procedure. Twenty-iour students at the University of Leidentook part in the experiment as volunteer paid subjects. The types of contextsentences and the idioms contained were rotated in such a way that each subjectreceived two sentences of each type and a subject would never see an idiomaticphrase in two different contexts. The subjects sat in front of the display at adistance of about 70 cm, and received the sentence one word at the time bypressing a key. The words of the sentence were displayed on the screen from leftto right, and remained visible until the end of the sentence. Some of the sen-tences required two lines on the display.

Results. The results, mean inspection times for the critical word, averagedover the "idiomatic point of uniqueness" word and the following word, arepresented in Table 4.2. The analysis of variance yielded a very significant effectof familiarity [F(l,ll) = 8.12,p < .01. by subjects, and F(l, 11) = 5.32,p <.05 by items]. Inspection times at the point of idiom uniqueness were virtuallythe same for familiar items and for control words, which were part of non-idiomatic phrases. Unfamiliar idioms, on the other hand, required significantlylonger inspection times.

The interaction between familiarity and transparency was significant, [F(l,23) = 8.39, p < .011 but this effect was mainly due to the unfamiliar items.Opaque unfamiliar idioms require significantly more inspection time than dotransparent ones [F(l, 5) = 6.79, p '( .05]. The interaction between type ofcontext sentence and familiarity was significant only for the unfamiliar idiomaticphrases. As compared with the latencies for the control words, reading a phrase

TABLE 4.2Mean Inspection Times (ms) Per Word for Familiar

and Unfamiliar Idiomatic Phrases in Idiomatic and Neutral Context,Averaged Over the Word That Was Judged as Point of Idiom

Uniqueness and the Word Following It

Idioms

Familiar Unfamiliar

Context Sentence Transp. Opaque Transp. Opaque Control

Idiomatic 423 435 527 549 417Neutral 432 441 556 586


containing a familiar idiom does not slow down the reading rate. So, neutralcontexts induce longer inspection times, but more so for opaque idioms.

Discussion. When a reader encounters the critical word of an idiom-thepoint of idiomaticity-she or he slows down with respect to a control in anequivalent string word only if the idiom is unfamiliar, and more so with a neutralcontext. This suggests that encountering a familiar idiom does not require anyadditional computation, and that processing goes on in the same way as with asentence containing a nonidiomatic phrase of the same structural complexity ofthe idiomatic one. On the other hand, when the critical, idiomatic word of anunfamiliar idiom is met, the reader slows down his or her reading pace, thusindicating the need for assigning an alternative interpretation to the sentence atthe point of idiomaticity.

Thus, the data show that readers, not surprisingly, have more difficulty ininterpreting unknown idioms, whereas they do not seem to have any difficulty ininterpreting a known idiom more than any other part of the sentence. When thecontext biases the idiomatic meaning, an unfamiliar idiom is understood better,but still is more difficult than a well-known idiom.

To conclude, we have obtained evidence for the notion that familiar idioms areprocessed without any problem. On the other hand, nonfamiliar idioms, es-pecially when opaque and especially when embedded in a neutral context, dooffer some processing problems. It seems reasonable to argue that this material iscomprehended in its literal sense up to the point when this is no longer possible.At this point, some extra computation seems to be required.

Thus the data here reported bring us to the following conclusion. Familiaridiomatic phrases are understood directly, without any additional effort, andprobably without the need of computing the literal interpretation. This, however,does not seem to be the case for nonfamiliar idioms, for which additional com-putational work seems to be required, and this especially when the context doesnot give any cue as to the presence of an idiomatic meaning of a critical string.

Study 4: Semantic Interpretation of Familiarand Unfamiliar Idioms

In this study, we presented the subjects with a number of idioms and requestedthat they choose, out of four alternatives, the phrase that correctly paraphrasedthe meaning of the idioms.

Material. Forty-eight idioms, chosen in such a way as to cover the wholerange of both the knowledge and the familiarity scales as determined in Study 1,constituted the material of this study. For each idiomatic phrase four alternativeparaphrases were made. One of these alternatives corresponded to the correct(dictionary definition) interpretation of the idiom; the others were chosen from

93

94 FLORES D'ARCAIS

the erroneous definitions produced by the subjects of the study to be reported inthe next section, or, in a few cases, created by the experimenter and two col-leagues.

Subjects and Procedure. Fifty students of the University of Leiden weregiven a booklet containing the 48 idioms with four alternatives each. The taskwas to choose the paraphrase that best corresponded in meaning to the meaningof the idiomatic phrase.

Results. For highly familiar idioms, the proportion of correct choices wasvery high, (.87) whereas for low-familiarity idioms the proportion of correctchoices was much lower, (.39) but still significantly different from chance (p <.001). The correlation between the proportion of choices of the correct alter-native and the knowledge score of the idioms, obtained in Study 1, turned outhigh, r = .85, p < .001.

Discussion. The subjects of this judgement study were in most cases capa-ble of selecting the appropriate paraphrase for familiar idioms. On the otherhand, they also performed better than by chance, even with unfamiliar idioms.

It is not surprising that familiarity with an idiom correlates highly with thecorrect interpretation. The fact that unknown idioms are assigned a correct para-phrase in a better-than-chance fashion suggests that people are capable of usingthe semantic information present in the string in order to reach a solution. Whatkind of strategies or principles do people use in searching for an interpretation ofan unknown and opaque idiomatic phrase? This question was at the basis of thefollowing study.

Study 5: Production of Paraphrases for UnknownIdiomatic Phrases

One of the questions asked in the present study concerned the semantic in-terpretation of unknown or poorly known idioms. What kind of interpretation canpeople provide when they meet an unknown idiom, or one that is very unfamiliaror difficult to remember? What kind of principles do people use in trying to makesense out of these idioms?

In the experiment here described we gave our subjects a number of unfamiliaridioms, with the request to provide in each case a paraphrase. We then analyzedthe paraphrases trying to isolate the principles that our subjects might have usedin producing them.

Material. Sixty-four idiomatic phrases, 32 familiar and 32 unfamiliar ones(as indexed by the results of Study I previously reported), constituted the mate-rial of the present study. Within each of these two categories, 16 were rather


transparent and 16 were rather opaque idioms. Together with this material, a totalof 66 filler sentences were presented, some having a metaphoric figurative mean-ing, some only a literal meaning.

Subjects. The subjects were 80 students of the University of Leiden whoparticipated in this study in groups, and performed the task required togetherwith another task, unrelated to the present one.

Procedure. The idioms were printed (without context) on a booklet withappropriate space for the subject to provide an answer. Each page contained 10phrases, and the order of the pages of the booklet was varied to ensure at leastsome randomization of the material.

Classification of the Responses. The responses were evaluated first on thebasis of their correctness. The evaluation consisted of matching the paraphrasewith the dictionary meaning. This work was done by four judges, and in mostcases the evaluation did not present any problem. The incorrect interpretationswere classified independently by the four judges on the basis of a number ofcategories created by the judges themselves while sorting the responses. Thesecategories then were pooled and after discussion given one single label. Theparaphrases then were sorted and assigned to one of these categories. Whenmultiple classification was possible, as in a few cases, the paraphrase given wasdiscussed among the judges and the classification was based on a forced choicebetween the two possible categories. Attribution of the paraphrases to the catego-ries thus created was in most cases completely unproblematic. In a few casessome discussion was required and in about 4% of the cases the paraphrases werebased on very idiosyncratic principles, could not be classified clearly, or, finally,disagreement remained among the judges.

Results. The proportion of paraphrases that closely matched the meaning ofthe idiomatic phrases was very high for the familiar idioms (.91) and signifi-cantly lower (.34) (p < .001) for the unfamiliar ones. For the further analysis,only the responses to the unfamiliar idioms were taken into consideration. Theproportions of the different types of interpretations given are summarized inTable 4.3. The paraphrases given by the subjects can be taken to be based on oneof the following principles:

1. Analogy: The idiomatic phrase is interpreted by analogy to a knownidiom, which contains a word or a constituent of a familiar idiom. For example,consider the idiom op ziin duim fluiten (to whistle in his thumb, meaning "notbeing able to obtain what one wants"). A few subjects gave an interpretation byanalogy to the idiom uit ziin duim zuigen (to suck his thumb, which means "tofind out something").

95

96 FLORES D'ARCAIS

Table 4.3Proportions of the Different Types of Interpretations

Given to the Unfamiliar Idioms

Interpretation Made on the Basis Of:

Semantic Prop. Metaphoric Literal NonCorrect Analogy One Word Extension Meaning Others Class.

.34 .17 .18 .12 .11 .04 .04

2. Use of semantic properties of one of the words of the idiom, mostly one ofthe nouns, without considering the whole phrase. Take, for example, the idiomuit de paardepoten komen (to come from the legs of the horse, which is ratheropaque and means "to go away"). Some interpretations given were of the follow-ing type: (a) "to originate from the country, from the farm" (horse as a farminganimal); (b) "to run fast" (horse as a fast-running animal).

3. Metaphoric extension of the action or state described in the phrase: Con-sider as an example of this operation the idiom voor twee ankers liggen (to lie attwo anchors, meaning "to be well taker care of"). The interpretations classifiedunder this category were of the type (a.) "not being able to move away" (twoanchors keep you tight); or (b) "to have doubts, to be uncertain" (the existence oftwo identical alternatives makes the choice difficult).

4. Literal meaning: The interpretation given was close to the literal meaning.The subjects were not able to figure out any idiomatic interpretation.

5. Other: In this category we classified a number of responses that were veryidiosyncratic, in the sense that the subjects seemed to create them without refer-ence to any obvious principle, or gave a paraphrase that to all judges seemedcompletely arbitrary.

Discussion. When presented with an unfamiliar idiom, and asked to give aninterpretation, people seem capable of giving paraphrases that in several casesare appropriate and often closely match the conventional meaning of the idiom.The interpretations proposed are obtained through familiar idioms or a meta-phoric analysis. Semantic information, which constitutes the meaning of thelexical units that are part of the idiomatic phrase, is used to construct an oftenappropriate or plausible interpretation.

GENERAL DISCUSSION

The studies summarized in this chapter have tried to analyze some of the pro-cesses that people bring to the interpretation and comprehension of an idiomaticphrase. The evidence obtained allows the following conclusions:


First, the input seems to undergo full syntactic analysis: Even very familiaridiomatic phrases, which in principle could be recognized after the first fewcomponent words, and do not require a syntactic analysis in order to be under-stood, seem to be parsed completely, much as any other linguistic string forwhich syntactic analysis is the normal and necessary route to obtain an interpreta-ble structure.

Second, the results of one of the experiments reported indicated that there is aclear difference in processing between well-known familiar idioms and new,unfamiliar ones. Although the first normally are processed without any addi-tional effort, the comprehension of an unfamiliar idiomatic phrase may requiresome additional computational effort.

Third, an unknown idiomatic phrase is given an interpretation based on anumber of principles. Some of these have been isolated as evidence for theprocesses that bring one to a meaningful interpretation.

In summary, it can be concluded that processing idiomatic phrases may pre-sent some computational problems only when the idioms are very or completelyunfamiliar. The interpretations given to these idioms are very often appropriateor at least close to their conventional meaning. In order to reach such interpreta-tions, the language user adopts a number of strategies that are based on simpleprinciples, some of which have been put in evidence by the work here described.

ACKNOWLEDGEMENTS

Part of the work here summarized has been supported by a grant of the C.N.R. ofItaly to the author.

REFERENCES

Bobrow, S., & Bell, B. (1973). On catching on to idiomatic expressions. Memory & Cognition, 1,343-346.

Cacciari, C., & Glucksberg, S. (1991). Understanding idiomatic expressions: The contribution ofword meanings. In G. B. Simpson (Ed.), Understanding word and sentence (pp. 217-240).Amsterdam, Netherlands: North Holland.


Ferreira, F., & Clifton, C. (1986). The independence of syntactic processing. Journal of Memoryand Language, 25, 348-368.

Flores d'Arcais, G. B. (1978). The perception of complex sentences. In W. J. M. Levelt & G. B.Flores d'Arcais (Eds.), Studies in the perception of language (pp. 155-185). Chichester, En-gland: Wiley.

Flores d'Arcais, G. B. (1982). Automatic syntactic computation and use of semantic informationduring sentence comprehension. Psychological Research, 44, 231-242.

Flores d'Arcais, G. B. (in preparation). The comprehension of familiar and unfamiliar idioms.Manuscript in preparation.

98 FLORES D'ARCAIS

Fraser, B. (1970). Idioms within a transformational grammar. Foundations of Language, 6, 22-42.Frazier, L. (1987). Sentence processing: A tutorial review. In M. Coltheart (Ed.), Attention and

performance XI1: The psychology of reading (pp. 559-586). Hillsdale, NJ: Lawrence ErlbaumAssociates.

Gibbs, R. W. (1984). Literal meaning and psychological theory. Cognitive Science, 8, 275-304.Gibbs, R. W. (1986). Skating on thin ice: Literal meaning and understanding idioms in conversa-

tion. Discourse Processes, 9, 17-30.Rayner, K., Carlson, M., & Frazier, L. (1983). The interaction of syntax and semantics during

sentence processing: Eye movements in the analysis of semantically biased sentences. Journal ofVerbal Learning and Verbal Behavior, 22, 358-374.

Schweigert, W. A. (1986). The comprehension (of familiar and less familiar idioms. Journal ofPsycholinguistic Research, 15, 33-45.

Swinney, D. A., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journalof Verbal Learning and Verbal Behavior, 18 523-534.

Tabossi, P., & Cacciari, C. (1988). Context effects in the comprehension of idioms. In CognitiveScience Society (Eds.), Proceedings of the Tenth Annual Conference of the Cognitive ScienceSociety (pp. 90-96). Hillsdale, NJ: Lawrence Erlbaum Associates.

Weinreich, V. (1966). Problems in the analysis of idioms. In J. Puhivel (Ed.), Substance andstructure of language (pp. 23-81). Los Angeles: University of California.

I I ACQUISITION ANDPROCESSING OF IDIOMS

99

The Acquisition of Idioms and5% the Development of FigurativeCompetence

M. Chiara LevoratoUniversity di Padova, Italy

We can play the game too, and drop our cover. I agree that there'snothing under sun so awful as a cliche, since, love it or leave it, it's all

the same cup of tea.-Umberto Eco, La bustina di Minerva, L'Espresso

LITERAL AND FIGURATIVE LANGUAGE

In the present chapter, which investigates idioms from a developmental perspec-tive, it is assumed that the ability to comprehend and produce idiomatic ex-pressions is inseparable from the development of figurative language. Let methus begin with an analysis of some of the problems inherent in the very defini-tion of figurative language and of the criteria that differentiate it from literallanguage. This analysis is not meant to be complete or detailed, but serves tointroduce a thesis that is central to this work: That is that the development ofchildren's ability to produce and understand idioms depends on the developmentof the same linguistic abilities on which figurative language as well as languagein general are based.

There is much controversy in psycholinguistics as to the validity of the dis-tinction between figurative and literal language (Dascal, 1987; Gibbs, 1984,1989). One of the weak points of this debate is that the criteria that are used todifferentiate literal and figurative language have not yet been identified defini-tively. I limit myself to an analysis of three characteristics of figurative languagethat are important for its acquisition. The first characteristic is a gap between thespeaker's words and his or her communicative intentions. A typical example of

101

102 LEVORATO

this is irony, where the intended meaning can be the exact opposite of the explicitmeaning. Another criterion that could distinguish literal from figurative languageis the latter's conventionality: It departs from its original meaning, the literal one,and acquires new meaning by means of strongly held conventions. Idioms areperfect examples of this, because according to convention the meaning of anidiom is idiomatic, not literal. Clear evidence of this can be found in researchwhere it has been proven that subjects took less time to process the figurativemeaning of an idiom than the literal meaning (Swinney & Cutler, 1979), or thatin any case recognition of the idiomatic meaning came first (Gibbs, 1980; Gibbs& Gonzales, 1985; Gibbs & Nayak, 1989; Gibbs, Nayak, Bolton, & Keppel,1988). The third difference between literal and figurative language is that figur-ative language is generally more dependent on the context than literal languageis. Actually, figurative expressions derive their meaning from context to anextent that varies with the degree of the conventionality of the expression.

There is no doubt these three characteristics play a crucial part in a child'sacquisition of figurative language, though they each may be debated on severalgrounds. First, concerning the say-mean distinction, it should be noted thatspeakers usually do not communicate explicitly what they really mean: In anycommunicative act linguistic production is based to a great extent on the assump-tion that the listener will make all the inferences necessary in order to get at themeaning intended by the speaker. Second, concerning conventionality, in numer-ous linguistic forms that strictly speaking may not be considered figurative, themeaning is linked to culturally determined conventions, as for instance in thecase of indirect speech acts. As for the third characteristic, literal language alsocan be context-dependent, because it is difficult to conceive of any expressionexisting outside a linguistic and extralinguistic context. However, the criteria that1 have just discussed must be taken into account in studying the process of theacquisition of figurative language in children. In fact, children have difficultyunderstanding (a) that a distinction may exist between what is said and what ismeant, (b) that the conventional meaning may differ from the literal one, and (c)that one needs to make use of all the available contextual information (thelinguistic and nonlinguistic context as well as the relevant world knowledge) toidentify the exact meaning of an expression.

Assuming that, for the aforementioned reasons, as well as others that I do nothave the space to elaborate on here, the distinction between literal and figurativelanguage does have heuristic value and therefore should not be neglected, mynext question concerns the possible consequence of such a distinction. I hope todemonstrate the inaccuracy of any model for the acquisition of figurative lan-guage that considers this acquisition to be a special competence, and to show, onthe contrary, that this acquisition is linked very closely to the development ofcognitive processes in general. The prevalent position for some time has beenthat figurative language must involve different processing from that involved inliteral language. Even if the linguistic distinction between literal and figurative

5. THE DEVELOPMENT OF FIGURATIVE COMPETENCE

language may be valid (and even if the speaker may be aware of it at some level),the processes underlying the comprehension and production of each type oflanguage are not necessarily different. In the literature a number of reasons aregiven for this, to which I add the following:

1. The idea that different processing could be activated for different, al-though still linguistic, stimuli goes against the principle of economy on whichcognitive functioning is based.

2. Differentiated processing would require some kind of mechanism to deter-mine whether and when a given expression should be received in its literalmeaning, and therefore undergo a normal processing, or whether it requiresspecial processing, but such a mechanism seems highly implausible.

3. There is so much figurative language present in normal conversation that aspecial processing would have to be almost as frequent as so-called normalprocessing. (According to the estimates of Pollio, Barlow, Fine, and Pollio,1977, an average of about four figures of speech are produced in every minute ofspeech.)

4. Nonliteral language where a discrepancy exists between what is said andwhat is meant can take a great variety of forms, ranging from expressions that areclearly figurative (e.g., proverbs, metaphors, idioms, and similes) to expressionsin which the discrepancy is more subtle (such as irony and indirect speech acts).

5. It seems more appropriate to analyze the literal-figurative dimension interms of a "fuzzy" category with some cases that are more typical than others.

6. Figurative language also includes expressions that are not figurative inthemselves, but become figurative in a given linguistic context.

Taken together, these observations lead us first to limit the distinction betweenliteral and figurative language, and second, to emphasize the existence of manydifferent typologies. That both figurative and literal language include a wealth offorms is a fact that plays an important role in our model for the acquisition offigurative competence (Levorato & Cacciari, 1992) in which the ability to pro-cess figurative language occurs in parallel with and as a function of a moregeneral ability to process language.

That a connection does exist between the acquisition of figurative languageand general linguistic development is undeniable. My goal here is to put to gooduse some of the theoretical tools that have proven so productive in the study ofchildren's language for a better understanding of the phenomenon of figurativelanguage. The connection between the development of figurative language andmore general aspects of the child's development has begun to be recognized. Inhis fine book on metaphor and irony, Winner (1988), for example, proposed amodel for the development of these two forms of figurative language that is basedon children's acquisition of world knowledge and on the growth of the ability to

103

104 LEVORATO

make inferences about the communicative intentions of the speaker. Thoughidiomatic expressions are conventional and not creative, unlike metaphor andirony, I believe that they are not acquired by associating form with meaning, butrather are acquired together with the development of skills that allow the child tocomprehend language as a whole, that is to create a semantic representation ofthe linguistic information. These skills include coding, making inferences, ac-tivating world knowledge, using imagination and creativity, finding out the com-municative intention of the speaker, activating metalinguistic knowledge andknowledge relating to the different kinds of discourse or text, and so on. In thecourse of the child's development, these processes and skills are ever moresubject to a tendency or need to search for the greatest possible degree ofcoherence among all the linguistic and nonlinguistic information processed at agiven moment and all other relevant information and previous knowledge.

With this, we touch on a crucial aspect of the acquisition of the ability to dealwith figurative language, which we can define as figurative competence. Figur-ative competence is not acquired all at once, but is pieced together in the courseof linguistic development and only can be acquired in full by a thoroughlycompetent speaker. In its most mature form it involves various linguistic skills,including: (a) the gradual broadening of word meaning, its position in a givensemantic domain, and its paradigmatic and sintagmatic relations; (b) the abilityto understand the dominant, peripheral, and polisemous meanings of a word, andalso the ability to perceive the relationship between a given meaning and otherrelated meanings; (c) the ability to suspend a purely referential strategy; (d) theability to understand the figurative uses of a word and the relationship betweenthe literal meaning and the figurative meaning; (e) the ability to process largeamounts of language, such as a text or a dialogue sequence, in order to identifythe meaning of ambiguous or unknown expressions; and (f) the ability to usefigurative language productively in the creation of new figures of speech bymeans of the lexical and syntactic transformation of preexisting figures ofspeech.

I present here a very rough outline of the role that the skills and abilities listedearlier play in the development of figurative competence, particularly in thecomprehension of idioms. In the conclusion of this chapter, after a report of theresults of several experiments, I provide a more complete analysis of the pro-cesses and mechanisms through which figurative competence is acquired.

IDIOM COMPREHENSION

The study of the psychological processes involved in an adult's or a competentlistener's use of idiomatic expressions is a radically different matter from study-ing the developmental processes involved in the acquisition of the necessaryskills for such use. The reason for this difference lies in the very nature of


idioms. Although these expressions originally described a particular situationliterally, their use has been extended to a greater and ever more varied range ofsituations and they have lost their literal character, which has caused them tobecome highly conventionalized. The frozen quality of such expressions resultsin their being processed by adults, or by those who have acquired them, pri-marily through their figurative meaning. It could be said, intuitively, that becauseto kick the bucket means "to die," no one would ever think, in the appropriatecontext, that it might mean "to kick the pail." Nevertheless, idiomatic ex-pressions often have a "literal" meaning that is different from the figurativemeaning, and it is just this fact of having two meanings that reinforces thetendency in children to go no further than the processing of the literal meaning.The literal strategy, which consists of applying only lexical and morphosyntacticcompetence to the decoding of the literal meaning, is not only the preferredstrategy and the one that is adopted first, but, especially in very young children,is the only strategy available.

Expressions that are relatively context-free for adults, such as idioms, can beprocessed correctly by children to the extent to which they can put to use all thecontextual information available for the recognition of the nonliteral meaning.

The few studies that have investigated the role of context among children areunanimous in finding that it does aid the comprehension of figurative meaning(Ackerman, 1982; Cacciari & Levorato, 1989; Nippold & Tarrant Martin, 1989).However, this finding generally has not been backed up by much theoreticalanalysis. One question in particular remains to be answered, and that is theproblem of establishing how and why an appropriate and sufficiently completecontext should lead to the comprehension of idiomatic expressions not previouslyknown to the child. I would like to try to answer this question with the followinghypothesis: Context operates if the child is able to take all the linguistic informa-tion surrounding the figurative expression and search it for those clues that mightlead to the idiomatic interpretation. Thus context provides background informa-tion for an hypothesis about the idiom's meaning, it aids the memory search forinformation that could serve to define the figurative meaning, and it provides thematerial for the appropriate inferential processes.

If we look at the role of context in adult comprehension of idioms we see thatit functions quite differently for children. Whereas adults retrieve the idiomaticmeaning first and only subsequently instantiate the interpretation of the idiomwith the linguistic context (Kemper, 1986) or, even if they integrate the idiomaticmeaning with the linguistic context parallel to the retrieval of the idiomaticmeaning, the integration does not depend on the contextual information, chil-dren, on the other hand, analyze the figurative expression at the same time as thelinguistic context and need the contextual information to grasp the idiomaticmeaning. The processes involved-from inference to the activation of the re-quired world knowledge-though not specific to the processing of idiomaticexpressions, are especially necessary in the comprehension of idioms, because

105

106 LEVORATO

the principle of compositionality can not be applied in the same simple and linearway as for literal language (Fraser, 1970; Katz & Postal, 1963).

As already mentioned, one important element of figurative competence is theability to use all the linguistic information available in the context in order toform a coherent semantic representation in which the semantic field establishedby the context will be consistent with the meaning given to the idiom. Theprimary characteristic of any type of text or discourse is coherence, because thatis what determines its status as a text or discourse rather than a simple phrasalsequence. The comprehension process, which is based on the assumption thatwhat is read or heard must be coherent, has its principle object in the identifica-tion of the main point of a text or discourse, and the creation of a semanticrepresentation that will mirror the coherence of the text. In the comprehension ofidioms, coherence is a key factor. The introduction of an idiomatic expressionabruptly changes the subject of a text, and if the interpreter goes no further than aliteral analysis, the coherence of the text is in jeopardy. The interpreter then musttake the comprehension process a step further and submit the linguistic informa-tion to semantic processing that will allow for the reestablishment of coherence.In the more advanced stages of linguistic development the awareness of a seman-tic gap automatically may activate strategies, in particular the figurative strategy,for reestablishing coherence. However, there must be present a need for coher-ence in the text before the child will be able to arrive at this figurative strategy.

Comprehension should not be considered as an all-or-none process, but as aprocess that can occur at different depths or levels depending on various factors:the demands set by the task, the previous knowledge of the interpreter, his or hergoals, and the extent of the acquisition of the ability to process language-thefactor most relevant in a developmental study. A mere perception of some senseof the message is inadequate for a comprehension at a deeper level, involving aconsideration of ambiguities, inferential steps, and nuances of meaning (Mistler-Lachman, 1972). In the first developmental stages every word or short sequenceof words is submitted to a semantic/syntactic analysis, and the meanings thatresult from this analyses simply are jux taposed in a semantic representation thatconsists of a collection of barely related items. Later the child acquires the abilityto process linguistic data in ever more complex ways, so that the products of asemantic/syntactic analysis of short segments of text are subjected to furtheranalyses with a view to integrating all the partial meanings into a global semanticrepresentation (Levorato, 1988). One of the main features of this global represen-tation is the degree to which the different semantic elements cohere. Thus, if apassage containing figurative expressions undergoes a superficial processing, thesemantic representation created from it could include elements that do not cohereamong themselves. As the children develop, however, they begin to search forcoherence among the codified items of information, and acquire a greater abilityto subject information to complex processing. Though it is difficult to say whichcomes first-the search for coherence or the growth of an ability to perform


complex processing of linguistic material-it is possible, however, to hypothe-size that the need to find coherence within one's own mental contents, or amongthe products of one's own processing, develops in parallel with the ability tocarry out those analyses that allow for the creation of a coherent representation.

A STUDY OF THE ROLE OF CONTEXT

I now describe the experiments that test our model of comprehension of idioms asdependent on the ability to interpret their meaning within the linguistic context.Our hypothesis was that comprehension would be better when the children couldmake use of contextual information in order to figure out what the idiomaticexpression meant. In Experiment 1, I the children's comprehension of idioms wasstudied under two conditions: (a) with idioms embedded in a linguistic contextconsisting of a short narrative, and (b) with idioms presented alone and out ofcontext. Because previous studies had shown that children have difficulty inexplaining the meaning of a sentence, comprehension was evaluated by means ofa multiple-choice test (MCT).

Eight common idioms were selected, all of them having both a figurative andliteral meaning (e.g., to break the ice), and for each idiom a story was preparedfollowing a simple problem-solving narrative formed by a setting with charac-ters, an initial event, and a solution. The last sentence of the story contained theidiomatic expression. For each story an MCT was prepared in which a questionabout the meaning of the idiom was followed by three possible answers: (a) theidiomatic meaning (idiomatic answer), (b) a paraphrase of the literal meaning(literal answer), and (c) a sentence that expressed a meaning different from both(a) and (b) but plausible in the context (associate answer). Both (a) and (c) madesense according to the semantic content of the story, whereas the literal answerexpressed a semantic content not congruent with the story. Subjects were firstand fourth graders. The following is an example of a story and of an MCTpresented with it:

A little boy named Paul moved to another town. It was winter so he had tochange school. His mother suggested that he should try and get to know his newschoolmates. Once at school he lent them his crayons and that helped to break theice.

What did Paul do when he broke the ice?

(a) He made friends with his schoolmates.(b) He broke a piece of ice.(c) He told his mummy everything.

'For more details, see Cacciari & Levorato (1989).

107

108 LEVORATO

As Table 5. 1 shows, more idiomatic answers were chosen when the idiom waspresented in context, because in fact the linguistic content of the story facilitatedcomprehension of the idiom. Literal answers were chosen very rarely when theidiom was presented in a story context and were almost totally absent among theolder children. In the no-context condition the literal answers were much morefrequent, but it should be kept in mind that the literal answers were not incorrect,because the idioms used in this study did have a literal meaning. Older childrenchose more idiomatic answers than did younger ones. These results confirm thatthe linguistic context in which the idiom is embedded plays a significant role inthe acquisition of the ability to comprehend figurative meaning.

Associate answers were chosen more often by younger children, irrespectiveof the presence or absence of a context. Although further investigation aboutassociate answers is necessary, we can say tentatively that sometimes youngchildren are able to reject a literal interpretation but are not able to identify theappropriate idiomatic meaning.

The stories provided for context had the effect of showing that the literalinterpretation did not make sense at all and that it was incongruent with anyexpectation formed on the basis of previous information. Moreover, the contextprovided the necessary semantic background to interpret the idiom correctly. Onemight ask at this point what would happen if the linguistic context pointed lessobviously to a figurative interpretation and to some extent did justify a literalinterpretation. This was the focus of an experiment in which stories presented tochildren contained a reference to an object, event, or situation semanticallyrelated to the literal meaning of the idiom. The aim of the second experiment wasto make the literal interpretation less implausible (though of course the figurativeinterpretation was still the correct one), to determine to what extent childrenmight be fooled by this type of context and whether or not they would chooseliteral answers more frequently than in the first experiment. The hypothesis wasthat if context represents an important source of information for the interpreter inidentifying figurative meaning, a context that is not incoherent with a literalinterpretation would produce fewer idiomatic answers than a context in which theliteral interpretation does not make sense. For the sake of brevity, we refer to the

TABLE 5.1Percentage of Idiomatic, Literal, and Associate Answers in theIdiomatic Context Condition and in the No-Context Condition,

for the Two Age Groups

Idiomatic Context No Context

Age Idiomatic Literal Associate Idiomatic Literal Associate

7 yrs. 59,1 18,6 22,3 21,2 56,8 2210 yrs. 87,5 3,7 8,8 51,1 40,1 7,8


story used in the first experiment as the idiomatic context and those used inthe second experiment as the literal context.

Stories with the same eight idioms as in the first experiment were presented tochildren of the same age as those in the first experiment and followed by an MCTin which the idiomatic and literal answers were the same as for the first experi-ment, whereas the associate answers were appropriate to the new context.

Table 5.2 shows the percentage of idiomatic, literal, and associate answerschosen by children with the literal context condition. In this experiment, as in thefirst, the older children chose more idiomatic answers than did the younger ones,as was predictable. It is to be noted that even when a literal context was used,idiomatic answers were still more frequent than when the idiom was presentedcompletely out of context (see Table 5.1). The percentage of associate answersthat resulted in this experiment confirms the developmental trend of the previousexperiment: This type of answer was chosen more often by younger children thanby older ones. This result provides us with another indirect confirmation of therole of context in comprehension: As the percentage of children who choseassociate answers was equal under the no-context as well as the context condi-tions (both idiomatic and literal), we can presume that this choice is typical ofchildren who are not disposed to use the context to understand the idiom; that isthe younger ones.

If we look at the percentage of idiomatic answers for figurative expressionpresented in an idiomatic story as compared with a figurative expression in aliteral story, it becomes clear that the semantic content preceding the idiom doesplay a role in inducing figurative, that is, idiomatic answers. In fact, the percent-age of idiomatic answers is higher in the idiomatic context, and this difference isstatistically significant. In both linguistic contexts the children were able to someextent to use the global meaning of the story to choose the most plausible answer,though under the literal context condition it was more difficult for them to reject aliteral interpretation.

The two experiments described so far produced very clear and coherent resultsthat strongly support our hypothesis regarding the role of context. There are stillsome questions that remain unanswered. One might wonder legitimately, forinstance, whether the results of the MCT might not overestimate the real ability

TABLE 5.2Percentage of Idiomatic, Literal, and Associate Answers

in the Literal Context Condition, for the Two Age Groups

Literal Context

Age Idiomatic Literal Associate

7 yrs. 51,3 28,7 20lo yrs. 72,5 17,5 10

109

110 LEVORATO

of the children to understand figurative language. Moreover, this task does notgive any evidence of what interpretive strategies are used by children to under-stand the language addressed to them. The performance might differ if thechildren were presented with another kind of task. The ability to process figur-ative language is a complex competence involving many different skills that varywith the cognitive and linguistic demands of the situation. This was the issue weintended to address in the third experiment.

The same stories as in Experiment I were presented to the children but theidiom was incomplete, and the task was to complete the expression. The aim wasto study the strategies used by the children in a task of this nature, especiallywhen they did not yet possess the idiomatic lexicon necessary for producing theexact words of the idioms. Our hypothesis was that it would be possible todistinguish strategies determined by figurative competence even prior to thelexicalization of figurative expressions in their conventional form, and that sucha competence would be linked to a global processing of the entire text in whichthe expression is embedded.

The children tested in this experiment were third and fifth graders; they wereolder than those in the previous two experiments, because we assumed that aproduction type of task would be more difficult than would a comprehension task.

The stories used in the first experiment (idiomatic context) were modified bydeleting a part of the idiomatic expression, beginning with the second constituentafter the verbal phrase. For instance, in the case of the story of Paul who went toa new school, the last sentence was: "Once at school he lent them his crayonsand that helped to break the . . . ... so the completion of the narrative was notfully open to any possibility but only to a small set of possibilities.

The subjects were presented with the eight incomplete stories and were askedto complete them with the word or words they felt were most appropriate.

The children produced three types of answers, the first of which was anidiomatic completion in which the exact wording of the idiom was used.

The second type of answer was a literal completion in which the childrenmerely adapted the words of the story rather than performing semantic analysis.The completions in this group were mostly of two types. In the first type of literalcompletion, the child saw the objects mentioned in the preceding part of the textas possible referents to the idiom. For example, for the story presented earlier inthe discussion of Experiment 1, some children write "to break the crayon," withthe result that the text was not totally incoherent and the completion did makesome sense-in fact, crayons that are lent out might be broken-but thesecompletions were clearly not products of a deep semantic processing of a text asa whole. In a story about a boy who had trouble fitting in with a new group ofclassmates, the breaking of a crayon is irrelevant. The listener, on hearing a storywith such an ending, would judge it somewhat out of place because of the gapbetween the expectations created by the text and the contents of the last sentence.


In the second type of literal completion the child chose a concrete referent thatsuited the verbal form of the idiom and that in everyday, although literal, lan-guage was the most likely nominal sintagm for that verb. For example, with theidiom to fall from the clouds (the idiomatic meaning is "to be astounded") acommon completion was stairs, which is easily associated with the verb to fall.Moreover, in Italian stairs agrees with clouds in gender and number (both arefeminine) and therefore follows the same preposition linked with its article(dalle) that was the last word of the story before the ellipsis. As for the coher-ence, it is to be noted that in this example the existence of stairs are not altogetherimplausible on the ship of a captain who has lost his way (but would be less so ifhe were the captain of an airplane!). However, if we consider the general mean-ing of the text with this completion, the lack of proportion in this interpretationbetween the main goal of the character (to find his way again) and the result (hefalls down the stairs) becomes obvious.

Both types of literal completion confirm that the strategy used was that ofsearching in the lexicon of the story for clues that would lead to a completion,and that the words chosen for the completions represented concrete referents.Thus, the inability to perceive a figurative use of the language seems to beassociated with a lack of ability to understand the general meaning of a text andwith a highly concrete mode of thinking.

The third type of answer was figurative completions, those that gave appropri-ate expression to the meaning of the idiom, the action, or the mental state of thecharacter to whom the idiom referred, without the child's knowing the idiomaticexpression. This produced expressions not canonical in Italian, such as "to breakthe fear" or "to break the shyness" (instead of to break the ice) or "to fall fromsurprise" (instead of to fall from clouds). With these completions the childrenproduced metaphorical expressions and showed that they were not insensitive tothe figurative potential of the language. Furthermore, they were able to gobeyond a strictly literal use of the language in order to fulfill certain commu-nicative aims. These children understood that the expression to be completed hadto provide the information coherent with the global meaning of the text. Thus, itseems that the inferential process of looking for the appropriate completion isdetermined by the child's knowledge of the main features of the story, as forexample, in the story of the new school, where the relevant fact is whether theboy achieved his aim in lending his crayons. Here we see that what is involved isthe child's ability to process linguistic information.

It appears that the strategies of children who made literal completions andthose of children who made figurative ones can be differentiated not only by theirability to use figurative language, but also by their overall cognitive and lin-guistic skills, and in particular by their ability to make the appropriate inferencesfrom the available linguistic information, either by means of a complex semanticprocessing, or through the use of appropriate general knowledge.

ill

112 LEVORATO

TABLE 5.3Percentage of Idiomatic, Literal, and FigurativeCompletions Produced by 8 Years and 11 Years

of Age. (Total Does Not Reach 100%,Because of Some Noriclassifiable Answers)

Completion Task

Age Idiomatic Literal Figurative

8 yrs. 14 39.2 41.211 yrs. 40 26.9 26.2

Table 5.3 gives the percentages of the three types of completions. Idiomaticresponses were much more common among the older children, who had acquiredidiomatic expressions through their greater exposure to adult language and wereable to produce them when they were provided with the beginning of an idiom.Literal completions were more common, on the other hand, among young chil-dren, as were the figurative completions, which also made up a consistentnumber of the responses of the older children (about one fourth of the total).

The figurative completions provide us with proof of a consistent and systemat-ic strategy not evident in earlier studies and not disclosed by the experiments thatused the MCT. This strategy tends to disappear as the children develop and theyacquire the exact wording of the idiomrs. These figurative completions prove theexistence of a figurative competence that comes before the acquisition of thespecific idiomatic expressions and that (a) is the basis for the figurative use notonly of idiomatic expressions but of language in general, and (b) precedes theconventionalizing phase.

One last consideration regards the difference between the three experiments,particularly between the multiple-choice task and the completion task. The for-mer was useful for measuring the effect of independent variables such as the ageof the subject and the informativeness of the context; however, it revealed lessabout the strategies the children used spontaneously. The completion task provedto be suitable for eliciting the strategies that the children presumably also used innonexperimental situations. Although the children tested this way were a yearolder than the others, a comparison st ll can be made between the results of thetasks (see Tables 5.1 and 5.3). It is to be noted that the idiomatic responses weremuch more frequent with the MCT, a result that is in line with the findings ofother studies about the relationship between comprehension and production infigurative language (Ackerman, 1982; Gibbs, 1986; Honeck, Sowry, & Voegtle,1978; Prinz, 1983; Vosniadou & Ortony, 1983; Winner, Engel, & Gardner,1980). The phenomenon of a gap between comprehension and production hasbeen commented on often in the literature about linguistic development. Never-theless, because the comprehension and the production tasks were presented to


different groups of children with varying ages, their relationship is dealt withlater.

ARE IDIOMS ACQUIRED BY ROTE?

A study of the acquisition of idiomatic expressions that aims at drawing up adevelopmental model of their comprehension must take into account all thefactors involved. Thus, I now investigate in greater depth the comprehension andthe production of idioms in relation to one factor that appears to affect theperformance of children: the degree of their familiarity with an idiomatic ex-pression. In studying this variable, especially in comparison with and in relationto other variables, we can judge more accurately the validity of the hypothesisthat idioms may be acquired by rote. However, even if it were true that thefamiliarity was a factor in acquisition (as Schweigert, 1986, observed amongadults), some form of figurative competence would be necessary for an under-standing of the meaning of an idiom, given that context can convey so manyvariations in meaning.

In the following experiments, we compared the role of context with the role offamiliarity, defined as the frequency with which an idiom appears in the languageaddressed to children. 2 If the ability to understand the figurative meaning of anidiomatic expression does depend on the acquisition of the wider cognitive andlinguistic abilities discussed earlier, then, presumably, the frequency of an idiomwould not play an important role in the process of identifying the idiomaticmeaning. A confirmation of this hypothesis would mean that idioms cannot betreated like nondecomposable long words and cannot be learned by rote. How-ever, because we could not exclude that familiarity might have some effect,either in itself or in combination with other factors, the object of this study was tofind out if there were conditions under which the familiarity factor might operate,in what way this occurred, and whether familiarity interacted with other vari-ables.

Along with the degree of familiarity, the variables considered in this studywere the age of the subjects, the type of context surrounding the idiom (idiomaticcontext vs. literal context), and the type of task performed (comprehension vs.production).

A preliminary survey was conducted in order to establish how familiar thechildren were with certain idiomatic expressions. In this survey, elementaryschool teachers were asked to evaluate the frequency of occurrence of 85 differ-ent idiomatic expressions in the language the children encountered in schooltextbooks, in everyday conversations, and on television. On the basis of the

2 For more details, see Levorato and Cacciari (1992).

113

114 LEVORATO

teachers' evaluations, we chose for the experiment five of the most commonidioms and five of the least common idioms (from now on referred to as familiarand unfamiliar idioms).

Two stories were prepared for each idiom, each story ending with an idiomaticexpression that expressed the story's conclusion. In the first story, which we referto as the idiomatic context, the literal interpretation was inappropriate because ofthe incongruence between the semantic field of the linguistic context and thesemantic field of the idiom. In the second story, hereafter the literal context, aliteral interpretation also would have been possible because the literal meaningrelated to objects or events in the context, although the idiomatic interpretationwas still the most appropriate one. Three responses also were prepared for eachstory and presented in a multiple-choice task. They consisted of an idiomaticinterpretation, a paraphrase of the literal interpretation, and a response thatdiffered from the first two and was plausible in the context (associate answer).

The subjects of the experiment, children from the first and fourth grades,listened to the stories and then were asked to perform the multiple-choice task.

The results of this experiment are shown in Table 5.4. The age of the subjectshad an effect on the idiomatic responses, the older children choosing them moreoften than the younger children did. These responses were also more frequent inthe idiomatic context than in the literal context. Familiarity was not a significantfactor in the choice of idiomatic responses. From this we conclude that the choiceof an idiomatic response is determined by the child's ability to process thelinguistic information surrounding the idiom and to identify the most plausibleresponse in a given context. This ability depends on the age of the child, andtherefore on his cognitive and linguistic development.

In the case of the literal responses, with fewer literal responses in the idi-omatic context, it is clear that context played an important role. However, theliteral responses were more common with unfamiliar idioms. Looking at theresults for both idiomatic and literal responses as a whole, we find confirmation,especially among older children, of the role of the linguistic context in activatinga figurative strategy based on an analysis of the linguistic information presented

TABLE 5.4Percentage of Idiomatic, Literal, and Associate Choices,for Familiar and Unfamiliar Idioms, in Idiomatic (IC) and

Literal Context (LC), Irrespective of the Age Factor

Familiar Unfamiliar

/C LC /C LC

Idiomatic Answers 57 40 51 38Literal Answers 14 30 24 42Associate Answers 29 30 25 20


by the story. This explains why the idiomatic responses increase and the literalresponses decrease when the linguistic context is semantically incompatible witha literal interpretation of the idiom. However, the fact that the children could befooled by the literal context into choosing a not entirely inappropriate literalresponse implies that they had not completely acquired and systematically ap-plied the figurative strategy, especially in the case of unfamiliar idioms. Thoughwe do not have an adult sample available, we assume that when figurativecompetence is fully developed, the listeners choose the figurative interpretationwhenever possible, basing the choice on their belief that whoever produced themessage would not have intended to create any ambiguity and would not havemeant a figurative expression to be interpreted literally. There is evidence in theliterature that adults are influenced much the same way as the children in ourstudy. In a study that had reading time as its dependent variable (Kemper, 1986),an unfamiliar idiom in a context with little information was read more slowlythan a familiar idiom in an information-rich context. Thus, when the on-lineprocesses in adults are examined, familiarity tends to enhance the effect ofcontext. The same seems to happen with young children: In a literal context,unfamiliar idioms are interpreted literally more frequently than are familiar idi-oms. Further proof is needed, but the results of our research so far confirm thatthe frequency of occurrence of an idiomatic expression does have an impact, inthat those children who do not make use of a systematic figurative strategy fallmore easily into the trap of accepting the literal interpretation when the idiom isunknown to them than when they already have had some occasion to associate itwith a figurative meaning.

As regards the associate responses, the results of this study agree to a strikingextent with the results of the first study. These responses, that is, were influencedonly by the factor of age, in that they were more common among the youngerchildren. At this point we are fairly confident that the tendency to choose thesehybrid responses-which are not literal and are closer to a figurative interpreta-tion without being quite appropriate-does not relate to the child's familiaritywith the idiom or to whether or not the linguistic information surrounding theidiom tends to trick the subject into a literal interpretation. The only factoroperating for these choices is age, that is, the level of the child's linguistic andcognitive development. The most plausible interpretation here is that associateresponses were chosen by children who were beginning to use a figurativestrategy, but were not yet able to analyze the text and extract the semanticinterpretation most appropriate to the context. This interpretation establishes twophases in the acquisition of idioms. In the first phase, children can and do usenonliteral strategies, although they are not yet able to process all the linguisticinformation completely in order to arrive at the most coherent interpretation ofthe idiom-and therefore they tended to choose the associate answers in ourexperiments. In the second and later phase, children have acquired fully theability to process linguistic information and so are able to create a fully integrated

115

116 LEVORATO

and coherent semantic representation-typified by the idiomatic response in ourexperiment.

These points are taken up again later when we discuss the developmentalsequence and the cognitive and linguistic skills that lead to figurative competenceand make up its substratum. Now let us turn to some important points concerningthe relationship between the comprehension and the production of idiomaticexpressions. In the experiment described next, we investigated the relative ef-fects of age and familiarity in a kind of production task, that is, a completiontask, and because the subjects did both a multiple-choice and a completion taskwe were able to examine the relationship between comprehension and produc-tion.

The material used in the experiment consisted of the same stories used in thepreceding experiment, but with part of the idiom left incomplete-the secondelement after the verb. The subjects of the experiment were children attendingthe second and fifth grades of elementary school. Each child performed thecompletion task individually, and then a week later, after hearing the completestory again, did an MCT.

In the completion task the same three categories of responses-idiomatic,literal, and figurative-were produced by the children as in the previous experi-ment on production. Table 5.5 gives the percentages of the three different re-sponses at the two age levels and for the two degrees of familiarity. The numberof idiomatic completions differed significantly according to the degree of famil-iarity, as familiar idioms were completed correctly more often than were un-familiar idioms. They also differed according to age, because older childrenproduced more than younger children.

Literal completions occurred more frequently among the younger children andwere more common when the idioms were unfamiliar. Thus, idiomatic and literalcompletions were perfectly complementary, because they were both affected bythe same factors-age and familiarity-but in opposite ways.

Figurative completions were more common with unfamiliar idioms, and fur-thermore were produced more often by older children than by younger children,although only when the idioms were unfamiliar. When the idioms were familiar,figurative completions were less common for both age groups.

TABLE 5.5Percentage of Idiomatic, Literal, and Figurative Completions

for Familiar and Unfamiliar Idioms;, According to the Two Age Groups

Familiar Unfamiliar

Age Idiomatic Literal Figurative Idiomatic Literal Figurative

8 yrs. 62 22 16 3 65 32i1 yrs. 82 6 12 3 47 50


In conclusion, the frequency of occurrence of an idiomatic expression in thelanguage addressed to children, which did not prove to be a significant factor in apreceding experiment that evaluated comprehension, turned out to be a crucialfactor in a production task. This does not mean that children are able to useidiomatic expressions spontaneously in their conversation, but simply that theycan produce appropriate completions when part of a familiar idiom is presentedto them in context. With unfamiliar idioms younger children tended to produceliteral completions more often than did older children, and the older children, inturn, produced more figurative completions. This serves to confirm that as theydevelop, children become progressively more skilled at perceiving the figurativepossibilities of language. Though the results of the productionlike task are veryclear, further investigation is needed into the way in which frequency of useinteracts with other variables such as type of linguistic context, general level ofworld knowledge, and expertise in a given field.

As described earlier, a week afterward the children were given an MCT inwhich the story was presented with the complete idiom and the children had tochoose among the three responses: idiomatic, literal, and associate.

I do not take too much time to analyze these results because the comprehen-sion task has been discussed already in connection with the preceding experi-ment. Now, I focus on the differences between the children's performances in theproduction task and their performances in the comprehension task. The aim ofthis comparison was two-fold. First, I wanted to check our hypothesis as to thedifferent degrees of complexity of the two tasks, and second, I wanted to verifythe hypothesis that a production task helps children to comprehend a text at a fardeeper level than when they have done an MCT.

The first hypothesis was confirmed by the number of idiomatic responsesproduced by the children in the production as compared with those produced inthe comprehension tasks, and especially by those cases in which a nonidiomaticcompletion was followed by an idiomatic response. The fact that emerges mostclearly is that for both age groups and for both familiar and unfamiliar idioms, avery high percentage of children made the "transition" from nonidiomatic com-pletions to idiomatic responses in the MCT. This confirms that the MCT is aneasier task for the children. However, it also indicates that MCTs should be usedwith some caution and in conjunction with other tasks in experiments of thisnature, because they could overestimate the real figurative competence of chil-dren.

Our second hypothesis about the effect of a production task in inducingcomprehension strategies was tested by comparing the children's performancesin the MCT of the present experiment with the performances of the children inthe preceding experiment in which they had not been subjected to the completiontask.

Table 5.6 shows the percentages of the different responses (idiomatic, literal,and associate) of the children in both the preceding experiment (7 and 10 years

117

118 LEVORATO

TABLE 5.6Percentage of Idiomatic, Literal, and Associate Choices

in the MCT in the Previous Eqperiment (7 and 10 Years Old)and in the Present One (8 and 11 Years Old)

Age & Task idiomatic Literal Associate

7 yrs: only MCT 36.5 29.5 348 yrs: MCT + Completion 70.8 17.2 1210 yrs: only MCT 56.5 25.5 1811 yrs: MCT + Completion 80 14.5 5.5

old) and the present experiment (8 and 11 years old). The comparisons involvefamiliar and unfamiliar idioms in both literal and idiomatic context. The mostinteresting comparison concerns the 7-. 8-, and 10-year-old children. The 7- and10-year-olds took part only in the MCT, whereas the 8-year-olds also did thecompletion task. In comparing the 7- and 10-year-olds, a natural developmentaltendency can be observed in which the number of literal responses decreasedwith age and the number of idiomatic responses increased. However, if wecompare the 8- and 10-year-olds, this tendency is not present and an inverse trendmay even be detected. Not only did the 10-year-olds not perform better than the8-year-olds, but they actually produced an average of 18.7% fewer idiomaticresponses.

These comparisons lead us to the conclusion that children tend to producemany more idiomatic responses when they go from a production task to a com-prehension task. The very fact of having first done the production task made thechildren more inclined to choose idiomatic responses in the comprehension task.Thus, the children tended not only to choose more idiomatic responses than theywould have been able to produce themselves, but they also chose more idiomaticresponses when they had already had a chance to submit the linguistic informa-tion to the deep-level comprehension processing required by the completion task.Further research will have to verify whether or not it is simply the fact of beingpresented twice with the text that brings about such a marked improvement in thechildren's performances. The most likely hypothesis for this is that the comple-tion task, because it is more demanding, forces the children to try to create amental representation of the text that is both logical and coherent. In other words,they must search for some idea of what the story is about. In relation to age, andso to their level of linguistic development, the children may or may not carry outthe task well, identifying the correct idiomatic completion in the case of familiaridioms, or in the case of unfamiliar idioms producing figurative or literal comple-tions. In any case, in the course of the experiment the children seem to haveacquired a greater capacity for recognizing the appropriate response when it waspresented together with other, less appropriate responses. The interpretation thatbest accounts for all the results presented so far is that the children's ability to


carry out these experimental tasks correlates closely with their linguistic abilityin general and especially with the ability to elaborate semantic information andachieve the global coherence. The value of the results of the aforementionedcomparisons should not be underestimated, either in reference to a theory of thedevelopment of figurative competence or to experimental methodology. It seemsclear, in fact, that performing a more demanding task first and then a simpler taskafterward did have some effect on the results of the latter. Further research isneeded to ascertain how general this tendency is and whether analogous conse-quences would result if the easier task were to precede the more difficult task.

CONCLUSION: TOWARD A MODELFOR THE DEVELOPMENT OF FIGURATIVE

COMPETENCE

The results of the experiments reported previously are proof that the acquisitionof idioms by children is far from a simple matter of passively learning conven-tional expressions, but is, on the contrary, a process involving complex linguisticand cognitive skills. There seems to be little recognition of this fact among thosewho study the language of children to judge by the research on the subject, whichhas not been abundant up to now (Ackerman, 1982; Cacciari & Levorato, 1989;Gibbs, 1987; Lodge & Leach, 1975; Nippold & Tarrant Martin, 1989; Prinz,1983). In the model for the development of figurative competence that I propose,the acquisition of figurative competence is tied to the development of a wholeseries of linguistic skills that give the child an ever greater control over his or hercommunicative possibilities. This linguistic development could be said to startwith a nominal realist phase, when an object and its name are viewed as one andthe same thing, and it could be said to have concluded with metalinguisticcompetence, which involves reflection about language itself. Very young chil-dren, as we know, believe that language is motivated and not conventional; thatis to say, the name of an object derives from that object, is intrinsic to it, and doesnot come from any agreement between speakers of the language. For example, ifthe "sun" were to change its name, children believe that it would lose its heat atwell (Piaget, 1923). Osherson and Markman (1985) found that young childrenwill accept that the name of an object can be changed, but believe that theproperties of the object will change with the name. They also may believe thatthe name has the same properties as the object so that, for example, the name fora train must be long like a train. In terms of the development of figurativecompetence, this phase of total identification of an object with its name withoutthe mediation of meaning can be thought of as Level 0.

Later, the name ceases to be part of the object and begins to take on thecharacter of a symbolic substitute. One of the first steps of this evolution is theovercoming of nominal realism. Meaning becomes of primary importance be-

119

120 LEVORATO

cause the child is now aware that the name refers to a meaning. The acquisitionof this awareness is linked to the discovery that the same linguistic label can begiven to various referents (as is the case for polysemous linguistic units), and thatthe same referent can be designated by various linguistic labels. The cognitivesubstratum to this discovery is the growth of a conceptual system and the rapidspeeding up and expansion of categorization processes. This point in the child'sevolution, which could be termed Level 1, still is characterized by a prevailingtendency to apply a literal strategy in the processing of text or discourse.

This tendency to apply a literal strategy in comprehension can be brokendown into the following aspects: (a) a tendency to conceive of meaning as themere sum of the meanings of the component words, (b) a tendency to submit thelinguistic information to shallow processing, (c) a tendency to be misled bythe imaginative contents of figures of speech, and (d) a tendency to consider onlythe concrete elements of an expression. These tendencies reinforce each otherand keep the child to a primitive concept of language, which is overcome only asthe child proceeds along a parallel course of linguistic and cognitive develop-ment.

Level 2 of the development of figurative competence is characterized by theimportant fact that the child now can go beyond purely literal strategies and apurely referential and literal use of the language. In this phase we see context andinferential processing being used to arrive at meaning and coherence. Our experi-mental results about the role of context for the semantic interpretation of idiomsare crucial for an understanding of the interpretive strategies used by children atthis level. As discussed earlier, there is a greater tendency to select an idiomaticresponse when the idiom is presented in an idiomatic context than when it ispresented out of context. This indicates that at a certain age children are able toperceive the incongruency of a literal interpretation with the contextual informa-tion surrounding an idiom. When the context cannot assimilate a given ex-pression, the child then activates interpretative strategies that go beyond theliteral. This phase could be defined as the phase of suspended literalness, be-cause the child now is able to make use of information that would not have beenavailable to him or her from a simple literal analysis. This phase correspondswith the last phase of cognitive pendulums in the linguistic realization (Nelson &Nelson, 1978). After a period of primitive forms of overgeneralization or narrowinterpretation of language followed by a period in which names are used accord-ing to rigid and inflexible rules of strict association between name and referent,finally children learn to accept a more flexible and open approach in whichlinguistic labels are used to form analogies and metaphors. For example, theymight say "this is like a dog." Now the child realizes that a discrepancy can existbetween an expression and its meaning. The linguistic awareness thus acquired,although rudimentary, allows the child to perceive that such a discrepancy shouldnot be interpreted as a communicative error or as a semantic anomaly. Thisinduces the child to activate inferential processes that will resolve the anomaly


and reestablish semantic coherence. This phase also is characterized by theability to process semantic information, allowing the child to make use of thecontext as background information for hypotheses about the appropriate in-terpretation of the idiomatic expression. Thus, the child's arrival at Level 2 of thedevelopment of figurative competence-which our data indicate happens at ap-proximately age 7 or 8 years-is the consequence of general linguistic develop-ment, and particularly of the acquisition of the ability to deal with semanticinformation. This is not acquired all at once, but is a slow and gradual process.Older children are better able than younger children to make use of contextualinformation, and at the same developmental level, different contextual informa-tion can facilitate to a greater or lesser degree the identification of idiomaticinterpretation.

The next level, Level 3, is characterized by the children's discovery of thearbitrary nature of language, which then leads them to realize that language doesnot always need to be literal. Although a communicative event was what inducedthe child at Level 2 to activate strategies other than the literal, at Level 3 the childcould be said to have acquired a kind of general rule that tells him or her not todepend too much on the surface form of a linguistic expression for meaning. Thisimplies an ability to use language for the most diverse communicative purposes,and also means that, given a communicative need, the child is able to think upvarious linguistic ways to accomplish this, the literal approach figuring as butone of many options. The acquisition of figurative language, and with it theability to produce and comprehend metaphors, similes, proverbs, and idioms, isonly one of the consequences of this development. According to our interpreta-tion of the experiment data, children who produced figurative completions hadreached this phase. Their strategy, in fact, is characterized by a lack of interest inthe literal aspects of the linguistic information and by the consequent attributionof priority to the meaning rather than to the significant. Compared with thepreceding level, here there is a much more marked tendency to use the significantas a clue that might lead to the discovery of meaning. With the awareness ofnonliteral language possibilities, meaning has become such a crucial element forcommunication that it detaches itself from the significant, which only serves asan aid in the search for meaning. To this is added the realization that the commu-nicative intentions of the speaker are what count and that the speaker is free touse any means he or she wishes to express those intentions. In cognitive terms,the child's thinking at this point is losing the concreteness typical of childrenunder 9 or 10 years of age. What stands out here is the fact that the child nowrealizes that the relationship between word and referent, as well as that betweensignificant and meaning, is a relationship of one to many; that is, a givenmeaning can be communicated by many different expressions. From this thechild also may arrive at the realization that some expressions are more effectivethan others in communicating a given meaning.

Level 4, in fact, is the concrete realization of all the potential developed in the

121

122 LEVORATO

preceding three levels. With a sufficient system of knowledge the child can linkexpressions to information and concepts already acquired and so may acquireconventionalized expressions such as idioms, frozen metaphors, formulaic ex-pressions, and so on. Those children w ho produced idiomatic completions in ourstudies based their responses on their knowledge of familiar expressions. Be-cause the prerequisite for acquiring these expressions in their standard form is theability to identify the appropriate meaning in a given context, this level is seen asthe logical successor to Level 3. Regarding the variables studied in our experi-ments, it could be said that if Levels 2 and 3 are the most strongly influenced bythe context, Level 4 is the one most influenced by familiarity: The acquisition ofthe idioms depends on the extent to which they are encountered in appropriatecontexts. The acquisition of some idiomatic expressions is possible even beforethis level, if the child is exposed to them enough, but Level 4 normally andlogically comes afterward because Level 3 provides the meanings to which thefigurative expressions then can be applied.

Level 4 has a limitation, however, which is that expressions are understoodand produced as indivisible units, as formulas that are applied in a fixed way andthat cannot be modified according to circumstance. This holistic approach doesnot allow the child to analyze the parts of an expression.

The complete acquisition of figurative competence comes only with Level 5,which is characterized by the ability to reflect on the meaning of a figurativeexpression and on the relationship between the significant and the meaning. Thelinguistic awareness characteristic of this level could be spoken of as meta-comprehension: Language now can be "analyzed" in order to understand therelationship between communicative intentions and surface expression. The in-terpreter is able to reconstruct meaning by making semantic inferences about theelements of the idiom and by referring tc his or her background knowledge. Thestrategies for interpreting figurative meaning and the inferential processes can beemployed both with and without the aid of context.

At its most mature, this level represents the skills, as far as figurative lan-guage is concerned, of a truly competent speaker. Regarding idiomatic ex-pressions, these skills can be described as follows: the ability to break down anidiom into its component parts and to make semantic inferences about these; theability to comprehend idiomatic expressions even when they have been subjectedto lexical substitution or syntactic and lexical variations; and the ability to gener-ate new idioms by means of syntactic and lexical variations on existing idioms.In conclusion, at this level the interpreters' linguistic awareness makes it possiblefor them to use figurative language for diverse purposes in which the communi-cation depends on the awareness of a special, intimate register full of tacitreferences and allusions (e.g., an idiomatic expression spoken with ironic intent,or an idiom changed in order to give hyperbolic expression to a given idiomaticmeaning as in the expression to burn a candle at three ends). This new compe-tence allows for the productive use of figurative language for a number of various


figures of speech. Our data about Level 5 are still incomplete, but research isbeing undertaken to investigate the concept of idioms as creative linguistic formsthat can be modified, and for which the meaning of each of the constituent wordsin both the on-line comprehension and the interpretation is accessed and plays arole in the overall comprehension of the idiom (Cacciari & Glucksberg, 1990).

The development of figurative competence parallels the transition from alimited linguistic competence in which language is concrete, referential, literal,and characterized by nominal realism, to a complete linguistic competence inwhich language use is supported by metalinguistic competence. This evolution ischaracterized by succession of phases in which the significant and the meaningare given more or less importance. The first change is from a tendency to makeuse of real life in order to acquire language to a tendency to use language toamplify the world knowledge (especially the social world). We could define thischange as a shift from a search for the significant to a search for meaning (thoughat the beginning of this process words are not real significants so much assignals). In the beginning, in fact, the child uses linguistic labels to accompanyhis practical knowledge of the world. Thus, small children will typically askadults for the names of objects they encounter in their exploration of reality. Laterthose names lose their character as extensions of the things they denote, and anew phase begins in which the child is searching for what lies behind the names(or between the names and the things)-in other words, the meaning. This phasesees the creation of a semantic system that will allow the child to recognize thatan entity belongs to a conceptual category. Regarding figurative competence, thephases in which meaning has priority coincide with Levels 1, 2, and 3, which arecharacterized by the progressive development of interpretive strategies for pol-isemic, ambiguous, or figurative terms; the child at Level I applies very limitedand elementary inferential processes, which become gradually more complexuntil at Level 3 an awareness of nonliteral possibilities of language opens up newinterpretive possibilities in the search for an appropriate meaning. At Level 4,where the significant is conventionalized, and Level 5, where metalinguisticawareness is attained, the significant-and with it the surface forms of lan-guage-return to the forefront and have priority: The interpreter realizes theconventional and arbitrary nature of language and has acquired the ability toreflect on language and its uses. Interpretive strategies also are used along withsyntactic and lexical strategies, to alter the significant according to particularcommunicative needs and to establish the relationship between the significantand the meaning.

In my opinion, the acquisition of idioms, although it may appear to be a"trivial" phenomenon in linguistic development, is fully a part of that develop-ment. With respect to other forms of figurative language that are less conven-tionalized, such as metaphors, idioms are peculiarly suitable for the study of thedevelopment of figurative competence. Although metaphors can be masteredalready at Level 3 (when the use of the interpretive strategies is quite advanced),

123

124 LEVORATO

because there is an explicit relationship between the expression and the speaker'sintention, the mastering of idioms comes at a later stage because of their conven-tional nature.

The model presented here is characterized by a developmental sequence inwhich each level gives way to and at the same time prepares the way for thefollowing level, which in its turn incorporates all the abilities acquired at thepreceding levels. However, this model does not exclude some variations tothe sequence. An inversion is possible, especially regarding the temporal rela-tionship between Levels 3 and 4 due to specific learning experiences. A key roleis played therefore by the systems of knowledge that are available at each stageof development and that assist in the transition from one level to the next. It isalso possible that a child may be at different levels at the same time according tothe extent of his or her knowledge relating to a particular semantic domain andthe linguistic expressions connected with it. For example, a child who is in-terested in baseball may develop expertise in the use of expressions relating tobaseball much more quickly than a child who does not play baseball. This doesnot mean that the only factors are expertise and experience, because in factstructural constraints do exist that limit that range of variability within oneindividual and among individuals. A child obviously cannot reach Level 3 beforeshe or he is able to carry out inferential processes of some complexity. Nor can achild reach Level 5 if his or her thinking is still tied to the concrete.

Idioms, and consequently comprehension, may vary in complexity with re-spect to various factors. Two main aspects need to be considered: the cognitiveaspect and the linguistic aspect. The former involves the semantic content ofidiom, so that, for example, idioms that represent concrete actions are presum-ably less difficult to understand, at least for children, than idioms that representemotions or mental states, and among the latter those referring to basic emotionsdiffer from those referring to complex cognitive states. The linguistic aspect mayinclude the morphosyntactic or semantic elements. Only one study, that by Gibbs(1987), has examined the relationship between linguistic factors such as semantictransparency and syntactic frozenness and the ability of the child to understandand explain the meaning of idiomatic expressions. His results seem to indicatethat children are able to carry out inferential processes to make the meaning of atransparent idiom compatible with the semantic field of the context. This sug-gests that research is needed not only into the effects of context in itself, but alsoits effects in relation to other variables. Syntactic frozenness refers to the degreeto which an idiom will tolerate variations in its syntactic form, such as thetransformation of an active verb into a passive verb, the insertion of adverbs, andso on. The most important result of Gibbs' investigation was that this variableplayed a role in the comprehension of idioms only for the kindergarteners andfirst graders, but not for third and fourth graders. In fact, younger childrenunderstood syntactically frozen idioms better than flexible idioms. The reasonfor this, according to Gibbs, is that syntactically frozen idioms are always recog-


nized in their standard form and so are acquired before flexible idioms. To acertain extent this result confirms the importance of frequency of idioms in theiracquisition, as it may influence the extent to which the child perceives the need todeal appropriately with such problematic linguistic input.

Idioms have apparent literal meanings that do not make sense in the linguisticcontext in which they appear. If the children could verbalize in Gricean terms theproblem they encounter with the comprehension of idioms, they might ask them-selves, "How am I to resolve this obvious violation of the rule that languagemust be relevant, one of the most pervasive rules of linguistic communication?"The obvious inadequacy of a literal interpretation leads the child to treat thefigurative expression as if it were in quotation marks, sustained, that is, by acommunicative intention that we could call "so to speak," in which the speakerintends to communicate something other than the literal meaning. An idiomaticexpression creates a situation that could be compared in a certain sense withsymbolic play. In symbolic play a real object stands for something other thanitself, and the players assign to it a meaning that suits their intentions (Piaget,1951). Similarly, an idiom is treated as a piece of language that stands for somemeaning other than that on the surface. The relationship between significant andmeaning is not the conventional one, but a deviation from the conventional that islegitimized by the speaker's intentions.

The analogy between the "pretence" as it applies to symbolic play and the"so to speak" as it applies to idioms goes beyond the pragmatic. In the same waythat the relationship between significant and meaning in symbolic play does havesome motivation, because a certain formal similarity is necessary between theobject used as a symbolic substitute and that which it represents, so too foridioms a certain similarity exists, though it is not always immediately apparent,between the surface form-the significant-and the figurative meaning. There isa reason why the idiom to bring coals to Newcastle has come to mean what itdoes, and does not mean the same as an idiom such as to kick the bucket. So too,this latter idiom is appropriate to describe a quick death and not drawn-out deaththroes, because the act of kicking takes only an instant. Thus the literal meaningmay be more relevant in the comprehension of idioms than sometimes has beenthought, as numerous recent studies have observed (Cacciari & Glucksberg,1990; Cacciari & Tabossi, 1988; Gibbs, Nayak, & Cutting, 1989; Glucksberg,McGlone, & Cacciari, 1990). In this regard we note some interesting reasonsprovided by fifth graders when asked to explain certain idioms. For instance, forthe idiom to be in seventh heaven ("to be very happy"), an explanation was, "Ifheaven is as wonderful as everyone says, just think what seventh heaven must belike"; or for to break the ice a child said, "It's hard to break the ice and the moretime passes, the harder it gets"; or for to have your head among the clouds ("tobe absent-minded") an answer was, "You can't be up among the clouds and haveboth feet on the ground at the same time. . . . The ground is the real world,while the clouds are your fantasies."

125

126 LEVORATO

The child, of course, is not alone in facing the task of discovering the meaningof new expressions, because adults do participate in this process. But meaningnever can be created by the simple association of a proposition with particularsemantic information. This cannot be an automatic process for various reasons;first, and somewhat paradoxically, because the high degree of conventionality ofidiomatic expressions make them quite flexible semantically so that their mean-ing may vary with the context. From this point of view, idiomatic expressions canbe seen as labels that refer to meanings slightly different one from another. Indifferent contexts to break the ice might mean "to start a conversation," "to getover embarrassment," "to overcome suspicion," or "to start up a friendship." Itseems that idioms contain a semantic core to which particular semantic elementscan be associated more or less as accessories. The meaning of an idiom acquiresever more depth as it is perceived in different contexts: With the experience ofever more and different contexts the child is able to perceive the core informationand the accessory elements or those that vary with the context.

There is another reason why the acquisition of idiomatic meaning necessarilymust be a constructive process characterized by successive adaptations and cor-rections. Even if the child may learn the meaning by asking an adult, she or hestill must find out how and when to use it. Children may learn that to kick thebucket means "to die," but they have vet to understand that this expression maybe appropriate in referring to a cockroach on the floor and not to the teacher, norto the death of Napoleon, or even worse, to the death of a hero (for whom itmight be more appropriate to use the expression to draw his last breath), al-though it could be used in speaking of Nero, especially if one wanted to conveycontempt. As with all language, idioms can be part of either a narrow or limitedcode, or a complex code. Idiomatic expressions are not limited, in fact, to thelanguage of "popular" culture, but can be found in every sphere of human lifeand communication. Indeed, what makes their use so fascinating is the way theyinvolve the imagination, make abstract meanings more concrete, add a wealth ofmeaning to simple concepts, and finally, make the commonplace conversationmore interesting. Like poetic language. idioms express in few words what wouldrequire many more words to express in literal terms but, given their conventionalnature they do not require complex creative strategies, such as those needed forpoetic language. Nor can they be called neutral denotations, because in fact theybring about a mutual understanding between speaker and interpreter based ontheir shared participation in a communicative game.

REFERENCES

Ackerman, B. P. (1982). On comprehending idioms: Do children get the picture? Journal of Experi-mental Child Psychology, 33, 439-454.

5. THE DEVELOPMENT OF FIGURATIVE COMPETENCE 127

Cacciari, C., & Glucksberg, S. (1990). Understanding idiomatic expression: The contribution ofword meanings. In G. B. Simpson (Ed.), Understanding word and sentence (pp. 217-240).Amsterdam, Netherlands: Elsevier.

Cacciari, C., & Levorato, M. C. (1989). How children understand idioms in discourse. Journal ofChild Language, 16, 387-405.


Dascal, M. (1987). Defending literal meaning. Cognitive Science. 11, 259-281.Fraser, B. (1970). Idioms within a transformational grammar. Foundations of Language, 6, 122-

142.Gibbs, R. (1980). Spilling the beans on understanding and memory for idioms in context. Memory

and Cognition, 8, 148-146.Gibbs, R. (1984). Literal meaning and psychological theory. Cognitive Psychology, 8, 191-219.Gibbs, R. (1986). Skating on thin ice: Literal meaning and understanding idioms in conversation.

Discourse Processes, 9, 17-30.Gibbs, R. (1987). Linguistic factors in children's understanding of idioms. Journal of Child Lan-

guage, 14, 569-586.Gibbs, R. (1989). Understanding and literal meaning. Cognitive Science, 13, 243-251.Gibbs, R., & Gonzales, G. P. (1985). Syntactic frozenness in processing and remembering idioms.

Cognition, 20, 243-259.Gibbs, R., & Nayak, N. P. (1989). Psycholinguistic studies on the syntactic behavior of idioms.

Cognitive Psychology, 21, 100-138.Gibbs, R., Nayak, N. P., Bolton, J. L., & Keppel, M. (1988). Speakers' assumptions about the

lexical flexibility of idioms. Memory & Cognition, 17(1), 58-68.Gibbs, R., Nayak, N. P., & Cutting, C. (1989). How to kick the bucket and not decompose:

Analyzability and idiom processing. Journal of Memory and Language, 28, 576-593.Glucksberg, S., McGlone, M., & Cacciari, C. (1990, November). Understanding idioms: The

psychology of allusion. Paper presented at the 31st annual meeting of the Psychonomic Society,New Orleans.

Honeck, R. P., Sowry, B., & Voegtle, K. (1978). Proverbial understanding in a pictorial context.Child Development, 49, 327-331.

Katz, J. J., & Postal, P. M. (1963). Semantic interpretations of idioms and sentence containingthem. Quarterly Progress Report of the Research Laboratory of Electronics, MIT, 70, 275-282.

Kemper, S. (1986). Inferential processes in the comprehension of idioms. Metaphor and SymbolicActivity, 1, 43-56.

Levorato, M. C. (1988). Racconti, store e narrazioni. I processi di comprensione dei testi [Theprocesses of text comprehension]. Bologna, Italy: 11 Mulino.

Levorato, M. C., & Cacciari, C. (1992). Children's comprehension and production of idioms: Therole of context and familiarity. Journal of Child Language, 19, 415-433.

Lodge, L., & Leach, E. (1975). Children's acquisition of idioms in the English language. Journal ofSpeech and Hearing Research, 18, 521-529.

Mistler-Lachman, J. L. (1972). Levels of comprehension in processing of normal and ambiguoussentences. Journal of Verbal Learning and Verbal Behavior, 11, 614-623.

Nelson, K. E., & Nelson, K. (1978). Cognitive pendulums and their linguistic realization. In K. E.Nelson (Ed.), Children's language (Vol. 1, pp. 223-287). New York: Gardner.

Nippold, M. A., & Tarrant Martin, S. (1989). Idiom interpretation in isolation versus context. Adevelopmental study with adolescents. Journal of Speech and Hearing Research, 32, 58-66.

Osherson, D., & Markman, E. (1975). Language and the ability to evaluate contradictions andtautologies. Cognition, 3, 213-226.

Piaget, J. (1955). The language and thought of the child. New York: Meridian Books. (Originalwork published 1923).

128 LEVORATO

Piaget, J. (1951). Play, dreams and imitation in childhood. New York: Norton.Pollio, H., Barlow, J., Fine, H., & Pollio, 11. (1977). Psychology and the poetics of growth.

Hillsdale, NJ: Lawrence Erlbaum Associates.Prinz, P. M. (1983). The development of idiomatic meaning in children. Language and Speech, 3,

263-272.Schweigert, W. (1986). The comprehension of familiar and less familiar idioms. Journal of Psycho-

linguistic Research, 15, 33-45.Swinney, D. A., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journal

of Verbal Learning and Verbal Behavior, 18, 523-534.Vosniadou, S., & Ortony, A. (1983). The emergence of the literal-metaphorical-anomalous distinc-

tion in young children. Child Development, 54, 154-161.Winner, E. (1988). The point of words: Children's understanding of metaphor and irony. Cam-

bridge, MA: Harvard University Press.Winner, E., Engel, M., & Gardner, H. (1980). Misunderstanding metaphor: What's the problem?

Journal of Experimental Child Psychology, 30, 22-32.

111-Formedness and6 Transformability inPortuguese Idioms

Teresa Botelho da SilvaUniversity of Cambridge

Anne CutlerMRC Applied Psychology Unit

Idioms are usually defined by their property of semantic eccentricity; they aremeaningful strings whose meaning is not a direct function of the meanings oftheir components. In linguistics and psycholinguistics, the main problem posedby idioms is the necessity (in grammatical analysis and comprehension) of treat-ing the string as a unit rather than decomposing it into its parts. The syntacticbehavior of idioms has been important to this problem principally because ofvariability among idioms in the extent to which they maintain their idiomaticityunder syntactic transformation.

Treating an idiomatic string as a unitary item has loomed large as a problembecause so many idioms are, in principle at least, susceptible to a parallel literalreading. Buckets can be kicked, beans spilled, and ice broken in the real, as wellas the idiomatic, world. Indeed, much of the literature on the processing ofidioms has focused on the question of whether the literal meaning of an idiom isactivated when the idiom is encountered; this question only has meaning, ofcourse, for idioms with a literal counterpart.

Not all idioms have a literal counterpart, however. For semantic or syntacticreasons, a literal interpretation may be completely ruled out. We refer to suchidioms as ill-formed. Semantic ill-formedness usually arises because of vio-lations of selection restrictions among the idiom's components-it is difficult tosee how one could literally rack one's brains, for example. Syntactic ill-formed-ness usually violates subcategorization restrictions; thus, only in the idiom canone be in the know, because elsewhere in the language know functions solely as averb.

Perhaps because so much of the literature on idioms and idiom processingdeals (as usual) exclusively with English, syntactic and semantic ill-formedness

129

130 BOTELHO DA SILVA AND CUTLER

among idioms has received little attention. Indeed, idioms without a literalcounterpart are often dismissed as accidental exceptions-on the grounds, forinstance, that they contain a unique form (e.g., eke out), that they include formsthat are fossilized survivors of originally free forms (e.g., cast a slur on), or thatone of their components has lost one of its original meanings (e.g., trip the lightfantastic). Some approaches to idiomaticity have even claimed that such colloca-tions cannot be considered "proper idioms" exactly because they violate thecondition of ambiguity, which in these approaches is considered to be a basicrequisite of idioms. Thus, Weinreich stated that "ambiguity is an essential char-acteristic of true idioms" (1969, p. 44); he denied the status of idioms to colloca-tions such as cockles of the heart or spic and span because they contain uniqueforms. Similarly, Makkai (1973) called such units "pseudo-idioms."

We agree that spic and span and cockles of the heart do not make strongclaims to idiomaticity. But rack one's brains and in the know certainly do, andalthough they do not contain unique forms, by virtue of their ill-formedness theyare not ambiguous. Therefore, on an ambiguity criterion such as that proposed byWeinreich, these phrases would necessarily have to be excluded from the realmof idiomaticity in English. To be sure, idioms without a literal counterpart arerelatively uncommon in English. This is probably one reason why an ambiguitycriterion has seemed defensible. If the source of ill-formedness is indeed quiteoften the presence of a unique form, then identifying ill-formedness with uniqueforms can be a practical step-by eliminating the latter, one also more or lesseliminates the former. Another reason, of course, may be that dismissing theproblem of ill-formedness simply facilitates clean elegant models of representa-tion (e.g., Weinreich's polysemy-based model).

The problem is that ill-formed idioms are far more frequent in other lan-guages, for instance French (Boisset, 197,3) and German (Greciano, 1983). Theyare also, as we shall show, extremely frequent in Portuguese. In such languages,ill-fonmedness in idioms may much more often than not go beyond unique forms;thus, it can hardly be dismissed as irrelevant by decree.

ILL-FORMEDNESS IN PORTUGUESE IDIOMS

In this chapter, we discuss ill-formedness in Portuguese idioms, and describe amemory experiment in which this factor was manipulated. To establish the sepa-rate categories of semantic and syntactic ill-formedness, we present severalexamples of each. (For each idiom, we give the Portuguese base form, followedby a literal English translation, then a paraphrase of the meaning). Examples I to6 are idioms for which a literal counterpart cannot be found for semantic reasons:

1. Dar bocas ("to give mouths" = to utter silly remarks).2. Meter medo a urn susto ("to scare a fright" = to be very frightening).

6. IDIOM ILL-FORMEDNESS AND TRANSFORMABILITY

3. Mandar Li fava ("send to the broadbean" = to send about one's busi-ness).

4. Fazer trinta por uma linha ("to make 30 by one line" = to misbehave).

5. Fugir a boca para a verdade ("to run away the mouth to the truth" = tobe sincere without intention).

6. Armar urn pe' de vento ("to set a foot of wind" = to make a scene).

The ill-formedness exhibited by these idioms is of a semantic nature. ExampleI has a simple verb plus noun phrase (V+NP) structure; dar is a transitive verband bocas is assigned the role of direct object. If one cannot literally "givemouths," it is not because give and mouths are syntactically incompatible, butbecause the semantic reading of the noun is not productive in association with thesemantic reading of the verb.

Likewise, the violation in Example 2 is also semantic; the prepositional phrase(PP) a um susto is assigned the role of indirect object to a verb whose indirectobject should be [+Animate], but susto (fright) can only be [-Animate].

In Example 3, we have a different type of semantic violation-the verbmandar (to send), in the sense used here, requires a directional locative comple-ment that must be either [+Animate] or [+Place], as in mandar a av6 (send tothe grandmother) or mandar ao mercado (send to the market). Fava (broad bean)does not qualify as a semantically adequate locative complement. It is possible,of course, that the original phrase from which this idiom derives may have beenquite regular, something like mandar buscarfavas (send to fetch broadbeans).

In Example 4, fazer (to make, to do) is used with the odd Direct Object trintapor uma linha (thirty by one line) when it would require a [+Concrete] or a[+Activity] Direct Object. It could be argued that the object is actually linha(line), but then we would have a syntactic irregularity because the numeralqualifier 30 would demand a plural noun. In the present stage of the language,there is no such activity or object, but again it seems possible that it oncedesignated some type of parlor game which has vanished from the memory of thespeakers' community.

Example 5 presents a similar irregularity to that of Example 3. Fugir (to runaway) usually demands a directional locative complement which has to be[+Concrete], either [+Animate] or [+Place]. A verdade (the truth) is neither;added to this we have the oddity of a [-Volitive] subject accompanying a verbwhose meaning implies volition.

In Example 6, the irregularity is again in the Direct Object, which should be[+Concrete] but is not-although, of course, p,' (foot) could be a concrete nounwere it not qualified by de vento (of wind).

Thus, one can say that Portuguese offers varied examples of highly meta-phorical idioms without a literal counterpart, clearly unambiguous, and withinteresting semantic irregularities. Now consider the idioms in Examples 7

131


through 12, which present marked syntactic irregularity in comparison to stan-dard Portuguese syntax; Mira Mateus, Brito, Silva Duarte, & Hub Faria, 1983):

7. Levar X a certa ("to take X to the certain" = to trick X).

8. Fazer caixinha ("to make little box" = to keep secret).9. Cair em si ("to fall in oneself" -= to become aware).

10. Andar na boa vai ela ("to go in the good goes she" = to be out on aspree).

11. Ter p6 a X ("to have dust at X' = to dislike X intensely).

12. Estar-se nas tintas para X ("to be oneself [reflex.] in the inks for X"not to care about X).

Example 7 is an instance of categorial irregularity. The structure of the idiomcan be analyzed as:

13. V + X + PREP (LOC) + ART + ADJ

The verb levar is a transitive verb with a valency of three, usually taking adirectional locative complement [PREP(loc) + NP] or [PREP(loc) + S1. Thecombination present in the idiom, [PREP/ oc) + ADJ], is therefore ruled out. Itis in the adjective certa, used in its feminine singular form, that the irregularitylies, as this seems to be performing the role of N in an NP, to the point of beingpreceded by DET (the definite article a, which appears in obligatory contractionwith the PREP a), in its adequate inflexional form, singular feminine.

In Example 8, fazer caixinha, we have an example of an incomplete NP. Theverb fazer (make or do) is a transitive verb, and the NP fills the role of directobject. The irregularity concerns the constituents of the NP. In Portuguese, thepresence of DET before the N is compulsory in the NP if the N is in the singularand is a countable noun. The only exception to this rule occurs when the Ndesignates an activity and, therefore, changes from countable to uncountable,allowing Example 14 but not Example 15, which is therefore marked * to signifyunacceptability:

14. Eu detesto cinema ("I hate cinema").

15. *Ele comprou cinema ("He bought [a] cinema").

In Example 9, cair em si, the ill-formedness resides in the prepositionalcomplement. The verb cair is an intransitive verb of movement, which generallyrequires a directional locative. The structure of this idiom makes the preposi-tional complement em si play the role of a directional locative; but in the case ofthe verb cair (fall), the referent of the locative cannot be identical to the referentof the subject-so Example 16 would be unacceptable:


16. A pedra, caiu na pedra, ("the stone, fell in/on the stone,").

The PREP COMP (and its flexion) shows that the personal pronoun caseused-the ablative-establishes an anaphoric relation with the subject, thusviolating the requirements stated previously.

In Example 10, andar na boa vai eta, we again have an irregularity in theform of the prepositional complement. Andar generally takes a locative comple-ment. In the present idiom, the structure of which is V+PREP(LOC)+ AR-T+ADJC+V+PRO, the phrase na boa vai eta is assigned this locative role. Theproblem rests on the fact that the preposition heading the locative is contractedwith a DET, the definite article a (feminine singular), which should be preceding(as it is prepared to do by the agreement) an NP; but the NP is absent. This absentNP would contain the noun with which the adjective boa is prepared to agree(because it appears in its feminine singular). It cannot be claimed that eta, thepersonal pronoun (feminine singular), has taken the NP function in the new Snode; pronouns cannot be determined by articles, nor do they take adjectives. Itmay be the case that there was once an N in the Prep Complement and that itcame to be dropped, but it seems more likely that this is a case of combination ofa process of fossilization of a question eta vai na boa (vida)? (Is she leading thegood life?), which came to be used as an idiom with the subsequent dropping ofthe noun vida. This whole fossilized former question would then have taken thecharacteristics of a noun; that might explain why it cannot be pluralized (i.e.,why Example 17 is possible but not Example 18):

17. 0 Miguel e a Isabel andam sempre na boa vai ela ("Miguel and Isabelgo always in the good goes she" = Miguel and Isabel are always out on aspree).

18. *0 Miguel e a Isabel andam sempre nos bons vdo etes "*Miguel andIsabel go always in the good go they."

The idiom in Example 11, ter p6 a X, shows a very odd subcategorizationirregularity and has the interesting syntactic structure of its literal paraphrase.The verb ter (have) only takes an indirect object if its direct object is a sensenoun, as in the idiom's paraphrase Ter 6dio a X ("have hate at X"). In the idiom,the object is not a sense noun but a common noun (dust) with which the verbter remains a two argument verb, therefore not allowing an indirect object.An indirect object (X) is, nevertheless, obligatory and is headed by the pre-position a.

Finally, in Example 12, estar-se nas tintas para X, we see another case ofsubcategorization irregularity caused by the combination of a reflexive pronounwith the verb estar. The reflexive pronoun is generally used only with transitiveverbs in Portuguese, indicating that the action operates on the subject that per-forms it. The verb estar is used to indicate a temporary state of being in a place,

133


mood, or condition (as opposed to the other verb which is translated by theEnglish be, namely ser, which indicates a permanent state of existence); it is nota transitive verb, so a reflexive conjugation is ruled out. There are exceptions tothis rule, involving verbs that indicate the subject's participation in an action ofthe affective type: apaixonar-se (fall in love), zangar-se (get angry). Verbs likeficar (stay), ir (go), rir (laugh), and .sorrir (smile) can occasionally take a reflex-ive pronoun, in which case they also lake on a new semantic reading. But theverb estar is not in either group and only in this idiom is it used with a reflexive.The case of the apparent parallel idiom estar-se borrifando para X ("be sprin-kling oneself for X"), which means exactly the same, is in fact totally differentbecause it uses a nonstandard construction that translates into the English presentcontinuous.

These are only a few of many ill-formed idioms in Portuguese. Whereas inEnglish ill-formed idioms, particularly syntactically ill-formed idioms such as byand large or in the know, are not the rule, in Portuguese such cases are muchmore common.

THE PSYCHOLINGUISTIC IMPLICATIONSOF ILL-FORMEDNESS

The mental processes involved in understanding idioms are the subject of a hugebody of research, and rightly so, because the phenomenon of idiomaticity hasimportant implications for models of the perception and production of language.If comprehension of a sentence consists, in the default case, of retrieving themeanings of individual words and combining them according to their gram-matical relations, then any case in which the meaning of a string is not a directfunction of such operations is obviously theoretically problematic (i.e., interest-ing).

Psycholinguistic interest in idiomaticity has focused upon three main aspectsof the problem:

1. There is the issue of how idioms are represented in the mind, that is,whether they are stored as lexical chunks or processed as any other word string(e.g., Cacciari & Tabossi, 1988; Estill & Kemper, 1982; Swinney & Cutler,1979).

2. There is the issue of access, the mechanism that makes a speaker opt foran idiomatic interpretation rather than for a literal interpretation when both areavailable (e.g., Gibbs, 1980, 1986; Ortorty, Schallert, Reynolds, & Antos, 1978;Van Lancker & Canter, 1981).

3. There is the question of the processing of transformed versus un-transformed idioms (e.g., Gibbs & Gonzales, 1985; Gibbs & Nayak, 1989;


Reagan, 1987); this, of course, also addresses the first problem (i.e., storage ofidioms in the mind), because if idioms are stored as unitary lexical representa-tions and are retrieved as such, then the introduction of syntactic transformsshould increase processing demands and make the lexical representations lesspromptly available, whereas if idioms are processed as any other word string,there should be no systematic differences in the amount of processing needed fortransformed versus untransformed idioms.

Semantic and syntactic ill-formedness are clearly relevant to these issues. Onthe one hand, the selection of an idiomatic meaning obviously faces no competi-tion from a potential literal meaning for ill-formed idioms because, by definition,they have no literal counterpart. Therefore, they may serve as a useful controlcondition in investigations of how idiomatic meanings are accessed. On the otherhand, the possibility exists that ill-formedness itself might serve as a signal thatthe incoming string is an idiom. This, in turn, might lead to the counterintuitiveprediction that ill-formed idioms could actually, in some respects, be processedmore easily than well-formed idioms-at least in languages in which idiom ill-formedness is common.

Indeed, although ill-formedness among English idioms has hardly been stud-ied at all, there are indications of support for this prediction. Brannon (1975)compared English idioms with and without literal counterparts in several experi-ments in which subjects judged the grammaticality of sentences or judgedwhether or not two sentences had the same meaning. Among idioms without aliteral counterpart, Brannon included several that we would describe as syntac-tically ill-formed (e.g., out of whack, go bananas); she found that idioms with aliteral counterpart (e.g., hit the bottle, down in the dumps) took longer to processin these tasks than matched unambiguous sentences did, but idioms without aliteral counterpart were processed faster than matched unambiguous sentences.

On the other hand, some semantically ill-formed idioms were included inexperiments by Mueller and Gibbs (1987) and Gibbs and Nayak (1989), althoughtheir criteria for ill-formedness do not correspond exactly to the distinctions weoutlined previously. Mueller and Gibbs (1987) reported that such ill-formedidioms were harder to process than well-formed idioms; thus, the processingeffects of ill-formedness, in English at least, remain unclear.

Syntactic ill-formedness also interacts in an interesting way with the issue ofsyntactic transformability. In English, those idioms that are syntactically ill-formed, such as in the know, also tend to be syntactically frozen (Fraser, 1970);i.e., to allow no syntactic transforms; e.g., *The know was what he was in aftertalking with the boss; cf. The doghouse was what he was in once his wife foundout. In Portuguese, however, ill-formedness and transformability are not fullyconfounded. Although syntactically ill-formed idioms tend to be frozen to agreater extent than well-formed idioms (Botelho da Silva, 1989), this is notalways the case. Some ungrammatical idioms allow transformations; as an exam-

135


pie, ter p6 a X can allow relative clause embedding, as in O p6 que a Beatriz tema escola ndo tem razdo de ser ("the dust which Beatrice had at school had noreason to be"). If one can imagine a continuum of syntactic ill-formedness, atleast in listener judgements, it is reasonable to suppose that transformed ill-formed idioms are even further along that continuum than their untransformedbase forms, if only because of the very low frequency with which they mightoccur, hence their relative unfamiliarity to listeners (cf. Cutler, 1982; Reagan,1987).

Gibbs and Gonzales (1985) made the interesting claim that the difficulty inprocessing transformed idioms should make them more memorable and henceeasier to recall. This claim is based on their finding that in a cued-recall task,transformable idioms were recalled more accurately than frozen idioms; thisfinding contrasts interestingly with the results from a phrase-judgment task inwhich subjects made significantly faster acceptability judgments to the frozenidioms than to the transformable ones. If our speculation that ill-formednesscould produce greater ease of processing has any foundation, and if, further-more, transformability of an ill-formed idiom only acts to increase ill-formed-ness, then we might actually expect the pattern which Gibbs and Gonzales found(for well-formed idioms) to be reversed with ill-formed idioms: Transformedidioms might prove easier to process and hence harder to recall.

So far, these speculations are just that; we cannot as yet provide an empiricalconfirmation or disconfirmation of their validity. At this point, we merely wish todraw attention to the potential usefulness s of ill-formedness as a factor in psycho-linguistic investigations of idiomaticitv. III-formedness is not highly commonamong English idioms; this suggests that the relevant experiments should becarried out in other languages that allow the relevant contrasts and this, too, is adevelopment that we would welcome.

In the next section, we report an initial study of the recall of idioms, in whichwe manipulated both ill-formedness and transformability. We were interested notonly in the contribution of these factors to the probability of recall per se, but alsoin establishing exactly how transformed idioms are recalled (i.e., whether theyare, in fact, recalled in their transformed form or, perhaps, in their base formwithout the transforms which were applied to them).

MEMORY FOR TRANSFORMED IDIOMS:AN EXPERIMENT

The properties of Portuguese idioms allow a systematic test of whether ill-formedness affects the way an idiom is processed. When ill-formed idioms arecomparatively rare and may differ systematically from well-formed idioms instructural characteristics (e.g., length), it is hard to institute a controlled test ofthe effects of ill-formedness. However, when ill-formed idioms are common, this


variable can be added to the repertoire of idiom properties that have been sub-jected to psycholinguistic investigation.

In the experiment that we describe here, we also exploit a new methodologyfor the study of idiom processing. We assess the recall of idioms via an inci-dental-learning paradigm. Subjects were presented with the idiomatic (and con-trol) materials and asked to assess the acceptability of each string. After theyhad completed this task with the entire set of materials, they were given asurprise recall task. Our primary measure of interest was performance on thisrecall task.

The materials were 30 Portuguese idioms, 15 well-formed and 15 ill-formed(of which 10 were semantically ill-formed and 5 were syntactically ill-formed).Within each subset of 15, 10 were transformable; the other 5 were frozen idioms,and thus admitted of no transforms. All idioms were embedded in sentencecontexts; for the 20 transformable idioms (10 of each type), two versions of eachcontext were given-one where the idiom appeared untransformed, one where itappeared transformed. We constructed 20 nonidiomatic control sentences, 10well-formed and 10 ill-formed. Each sentence contained a proper name or a noun

TABLE 6.1Examples of Sentences Presented in the Experiment

Well-Formed III-Formed

Transformed No tribunal os pontos foram pos-Idioms tos nos is pelo juiz Monteiro.

(In court the dots were put onthe i's by Judge Monteiro.)

UntransformedIdioms

Frozen Idioms

Control Strings

O policia deu cabo do canastroao ladrao de carros.(The policeman destroyed thebasket to the car thief.)

No verao as criangcas passamsempre pelas brasas depoisdo almoco.(In summer, children alwayspass by burning coals afterlunch.)

O gerente chegou ao bancomuito atrasado para areuniao.(The manager arrived at thebank very late for the meet-ing.)

o p6 de vento que o Artur ar-mou nao teve razio de ser(The foot of wind which Ar-thur set had no reason tobe.)

o cabeleireiro tern p6 is cli-entas que nunca sabemcomo querem o cabelo.(The hairdresser has dust toclients who never knowhow they want their hair.)

Os motoristas estio-se nastintas para os pe6es.(Drivers are (reflex.) in theinks for pedestrians.)

O Pedro vai nunca pescarneste lago.(Pedro goes never fishing inthis lake.)

137


denoting a profession; these were intended for use as recall prompts. Table Icontains examples of the sentences presented to subjects, and the full set ofidioms tested appears in the Appendix.

The transforms that we used were the following: (a) relative clause embed-ding, (b) insertion, (c) active nominalization, (d) passivization, (e) clefting, and(f) permutation. These were not systematically manipulated; only one operationwas applied to each idiom. In a previous study (see Botelho da Silva, 1989), ahierarchy of operations, analogous to that established by Fraser (1970) for En-glish, was established for Portuguese; in this hierarchy, the operations we usedwere ordered as shown earlier, with relative-clause embedding producing thehighest proportion of "acceptable" judgments and permutation the lowest. In-terestingly, this study produced a contrast with Fraser's hierarchy, in that it wasnot necessarily the case that acceptability at a given level of the hierarchy impliedacceptability at lower levels. We note that Gibbs and Gonzales (1985) also foundthat mean acceptability ratings for transformed idioms produced by a group ofsubjects did not correspond to the intuitive hierarchy proposed by Fraser and didnot always pattern in a strictly hierarchical order. In the present study, wedeliberately chose transformed constructions across a range of acceptability be-cause of the nature of the subjects' ostensible primary task in the acquisitionphase of the experiment-acceptability judgment.

Two sets of materials were constructed, differing only in which version of atransformable idiom appeared in each set; for each such idiom, its untransformedversion appeared in one set and its transformed version in the other (with gram-maticality, of course, counterbalanced across sets).

Fourteen native speakers of Portuguese took part in the experiment; sevenreceived each materials set. The subjects were tested individually and the 50sentences of each set were randomized separately for each subject. Subjects readeach sentence aloud from a card and then judged it as "perfectly correct,""possible," or "impossible" in the Portuguese language. After the 50 sentenceshad been judged, the subjects were asked to write down as much as they couldrecall of the entire set of sentences. They were told to write down full sentenceswhere they could, but also any parts of sentences they could recall, and they weregiven unlimited time for this task. Recall prompts were given for (randomlychosen) sentences of which no part had been recalled (but a maximum of fourrecall prompts was given to each subject).

The results can be exploited in several dimensions. We have data on how oftena given idiom was recalled and the form in which it was recalled; we also havedata on how acceptable subjects judged the idiom to be in the form in which itwas presented. We can, therefore, look at recall per se as a function of gram-maticality and transformability. We can look at recall in the light of acceptability,and we can look at how idioms are recalled, again as a function of the variableswe manipulated.


Effects of III-Formedness and Transformabilityon Recall

For a measure of recall per se we counted any recognizable version of the targetstring, irrespective of syntactic form or paraphrase. The clearest finding fromthis analysis was that idioms were recalled more often than control strings-19.29% of idioms overall were recalled in comparison with 7.14% of controlstrings. This difference was significant [X2 (1) = 20.06; p < .001].

Surprisingly, however, grammaticality had no effect on probability of recall.There were 40 instances of recall of well-formed idioms (19.05%) and 41 ofrecall of ill-formed idioms (19.52%), an insignificant difference. Among the ill-formed idioms, there was again no significant difference in the probability ofrecall between the semantically and the syntactically ill-formed set.

Transformability also had no effect, with the probability of recall for frozenand transformable idioms virtually identical.

Transformation, in contrast, had a strong effect: The probability of recall foruntransformed idioms (37%) was almost twice as high as for transformed idioms(21%). This difference is significant [X2 (1) = 8.35; p < .01].

Acceptability and Recall

As in the preceding study (Botelho da Silva, 1989), the acceptability ratings fromthese subjects were lower for transforms of ill-formed idioms than for transformsof well-formed strings, in accordance with the tendency of ill-formed idiomstowards syntactic frozenness. There was, however, no discernable indication inthe results of a systematic relationship between recall probability and ratedacceptability.

How are Idioms Recalled?

When we examined the precise form in which subjects recalled the idioms, wefound the most surprising result of this study. When transformed idioms wererecalled (remember that this happened in only 21% of cases), they were moreoften than not recalled without the transforms which had been applied to them.Only in one third of such cases were they recalled as they had been heard,another one tenth were paraphrased, and the majority were recalled in their baseform.

This pattern held equally true for ill-formed idioms, both semantically andsyntactically, and for well-formed idioms.

The frequency of paraphrase in comparison with exact recall was low foruntransformed transformable idioms and for frozen idioms, and showed no rela-tion with well-formedness. By comparison, an interesting pattern was revealed in

139


recall of the control strings. More than half of recall instances of these involvedparaphrase, and paraphrase was much more likely for well-formed strings thanfor ill-formed strings. This presumably reflects the long-known fact that the formof linguistic material is less well recalled than the content. Consider, however,the fact that idioms resisted this tendency to a considerable extent (for all un-transformed idioms, only 20% of recall instances involved paraphrase, comparedwith over 50% of recall instances for control strings). We suggest that thisindicates closer connection between form and content in the case of idiom stringsand is just what would be expected if idioms are being processed as, in somesense, unitary items.

CONCLUSION

Our experiment constitutes only a pilot study but it offers some very interestingimplications for further research-although these are not necessarily the implica-tions we were expecting when we undertook the study! Of our two original aims,the first was to investigate the role of Il-formedness in idiom processing. Wefound none; ill-formed and well-formed idioms produced very similar recallresults. However, even this negative result has potential implications; if it provesreliable and parallels results from other processing tasks, it would argue againstprocessing models which predict different processing costs for idioms whichhave to be distinguished from a literal counterpart versus idioms which don't (seeCutler, 1983, for further discussion of this issue). Our second aim was to exploreincidental learning methodology as a tool for.investigating idiom processing; ourresults suggest that this methodology could prove extremely useful.

We hope others will follow up and extend our results on recall for idiomspresented in transformed form; our subjects tended to detransform such items andrecall the base form instead. We suggest that this finding is very consistent with aview of idioms as unitary lexical items, and, along with our finding that idiomswere more resistant to paraphrase than were control strings, lends support tolexical-unit models of idiom representation (e.g., Swinney & Cutler, 1979).

Finally, our failure to find advantages for transformable idioms in ease ofrecall contradicts Gibbs and Gonzales' (1985) prediction to this effect; indeed,our failure to find effects of well-fonnedness could also be construed as contraryto a prediction that idioms which require differential amounts of processing willproduce differential recall results. Note also that we found better recall foruntransformed idioms, which suggests that if the transforms we used increasedthe amount of processing necessary for understanding, then the increased pro-cessing certainly did not lead to increased recall probability.

Este estudo e s6 uma gota no oceano. mas talvez contribua para descobrir ofio da meada deste fen6meno linguistico que se tem provado urn osso bem durode roer.

REFERENCES

Boisset, J. H. (1978). Idioms as linguistic conventions. Unpublished doctoral dissertation, Univer-sity of Florida, Gainesville.

Botelho da Silva, T. (1989). Processing offrozen idioms. Unpublished master's dissertation, Cam-bridge University, Cambridge, England.

Brannon, L. L. (1975). On the understanding of idiomatic expressions. Unpublished doctoral dis-sertation, University of Texas, Austin.


Cutler, A. (1982). Idioms-The colder the older. Linguistic Inquiry, 13, 317-320.Cutler, A. (1983). Lexical complexity and sentence processing. In G. B. Flores D'Arcais & R. Y.

Jarvella (Eds.), The process of language understanding (pp. 43-79). Chichester: Wiley.Estill, R. B., & Kemper, S. (1982). Interpreting idioms. Journal of Psycholinguistic Research, 11,

559-568.Fraser, B. (1970). Idioms within a transformational grammar. Foundations of Language, 6, 22-42.Gibbs, R. W., Jr. (1980). Spilling the beans on understanding and memory for idioms in conversa-

tion. Memory and Cognition, 8, 149-156.Gibbs, R. W., Jr. (1986). Skating on thin ice: Literal meaning and understanding idioms in conver-

sation. Discourse Processes, 9, 17-30.Gibbs, R. W., Jr., & Gonzales, G. (1985). Syntactic frozenness in processing and remembering

idioms. Cognition, 20, 243-259.Gibbs, R. W., Jr., & Nayak, N. P. (1989). Psycholinguistic studies in the syntactic behavior of

idioms. Cognitive Psychology, 21, 100-138.Greciano, G. (1983). Signification et denotation en allemand-La semantique des expressions idi-

omatiques [Signification and denotation in German: The semantics of idiomatic expressions].Recherches Linguistiques, 9. University de Metz.

Makkai, A. (1973). Idiom structure in english. The Hague, Netherlands: Mouton.Mira Mateus, M. H., Brito, A. M., Silva Duarte, I., & Hub Faria, 1. (1983).Gramatica da Lingua

Portuguesa [Grammar of the Portuguese Language]. Coimbra: Almedina.Mueller, R., & Gibbs, R. W., Jr. (1987). Processing idioms with multiple meanings. Journal of

Psycholinguistic Research, 16, 63-81.Ortony, A., Schallert, D., Reynolds, R., & Antos, S. (1978). Interpreting metaphors and idioms:

Some effects of context on comprehension. Journal of Verbal Learning and Verbal Behavior, 17,465-478.

Reagan, R. T. (1987). The syntax of English idioms: Can the dog be put on? Journal of Psycho-linguistic Research, 16, 417-441.

Swinney, D. A., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journalof Verbal Learning and Verbal Behavior, 18, 523-534.

Van Lancker, D., & Canter, G. J. (1981). Idiomatic versus literal interpretations of ditropicallyambiguous sentences. Journal of Speech and Hearing Research, 24, 64-69.

Weinreich, U. (1969). Problems in the analysis of idioms. In J. Puhvel (Ed)., The substance andstructure of language (pp. 23-81). Berkeley: University of Califomia Press.

APPENDIX

Well-Formed Idioms

1. Bater a bota ("to beat the boot" = to die).

141


2. Andar de vento em popa ("to go with the wind in the stem" = to go verywell).

3. Passar pelas brasas ("to pass by the burning coals" = to take a nap).

4. Ir a nove ("to go at nine" == to move very quickly).5. Baixar a bolinha ("to lower the little ball" = to quiet down).

6. Estar nas suas sete quintas ("to be in one's seven farms" = to feel veryhappy).

7. P6r o carro a frente dos bois ("to put the cart before the oxen" = tochange the natural order of things).

8. Dar cabo do canastro de ("to destroy the basket to" = to spank).

9. P6r no prego ("to put in the nail" = to pawn).10. Apalpar o terreno ("to feel the ground" = to ponder before taking

action).11. Puxar a brasa a sua sardinha ("to pull the burning coal to one's sardine"

= to defend one's interests).12. P6r os pontos nos is ("to dot the i's" = to settle matters).

13. Dar o n6 ("to give the knot" = to get married).14. Perder a cabeca ("to lose the head" = to get out of control).15. Fazer uma fita ("to make a film" = to create a scene).

III-Formed Idioms

Syntactically Ill-Formed

1. Estar-se nas tintas ("to be [reflex.] in the inks" = to be indifferent).2. Cair em si ("to fall inside oneself" = to become aware).

3. Dar o dito pelo nao dito ("to give the said by the not said" = to changeone's position).

4. Ter p6 a X ("to have dust at X" = to hate X).5. Por-se a pau ("to put oneself at stick" - to be on the alert).

Semantically IIl-Formed

6. Dar a lingua ("to give to the tongue" = to chatter continuously).7. Esticar o pernil ("to stretch the spindle-shank" = to die).

8. Meter os pes pelas maos ("to put the feet through the hands" = toblunder).

9. Mandar a fava ("to send to the broadbean" = to send away, to dismiss).10. Armar um pe de vento ("to set a foot of wind" = to make a scene).11. Dar a luz ("to give to the light" = to give birth to).

6. IDIOM ILL-FORMEDNESS AND TRANSFORMABILITY 143

12. Dar a mao a palmat6ria ("to give the hand to the ferrule" = to recognizea mistake).

13. Dizer cobras e lagartos de ("to say snakes and lizards of" = to speak illof).

14. Fazer trinta por uma linha ("to make 30 by one line" = to misbehave).

15. Escangalhar-se a rir ("to destroy oneself laughing" = to laugh heartily).

7 The Activation of IdiomaticMeaning in Spoken LanguageComprehension

Patrizia Tabossi and Francesco ZardonUniversity of Ferrara and University of Bologna, Italy

As pointed out by most authors in this volume, idioms are multifaceted objects,whose study requires various viewpoints and different methodological ap-proaches. They are not only complex, but also in many respects elusive, to thepoint that it is probably a useless exercise to try and define them (Dascal, 1987;Fernando, 1978). Several aspects contribute to render the characterization ofthese expressions a challenging task. Typically, idioms are assumed to be stringsof words whose semantic interpretation cannot be derived compositionally fromthe interpretation of their parts. According to this characterization, however,break the ice is as good an idiom as switch off or by accident, even though ourintuition suggests different degrees of "idiomaticity" of these expressions.

Syntactic behavior does not show any higher homogeneity: It is well known,for instance, that idioms vary widely in "frozenness" (Fraser, 1970). Thus, laydown the law accepts most syntactic operations, whereas an expression of identi-cal syntactic structure such as blow off steam accepts none.

Similarly, although the semantic interpretation of some idioms appears to becompletely unrelated to the meanings of their component words (e.g., kick thebucket), the meanings of others can be inferred at least to some extent from theirelements (e.g., carry a torch). These differences probably underlie the seeminglycontradictory theoretical claims presented by current research on idioms. On theone hand, there are those who stress the aspects of semantic opacity of theseexpressions (Katz, 1973). On the other hand, an increasingly popular view positssome level of compositionality in the comprehension and production of idioms(Cacciari & Glucksberg, 1990; Gibbs & Nayak, 1989; Wasow, Sag, & Nunberg,1983).

Related to the issue of semantic opacity/transparency is the question of how

145

146 TABOSSI AND ZARDON

idioms fit in the distinction between literal and figurative language, provided thatthis distinction is tenable (Rumelhart, 1979). Here the difficulty arises from thefact that some idioms share numerous characteristics with figurative expressions,in particular with metaphors (Gibbs & O'Brien, 1990; Lakoff, 1986). Otheridioms, however, are so conventional and frozen to make it hard to appreciate thedifference between them and actual individual words: Leaving aside orthography,how do expressions such as nonetheless and by and large differ from each other?

Given these great diversities, it is not surprising that the scope of most studieson idioms is often rather limited, with theoretical claims and empirical findingstypically being restricted only to some subset of those expressions, or to somespecific aspects of the many processes involved in their comprehension (or both).The work presented here is no exception, and the admittedly partial approach ittakes is one that looks at idioms from the standpoint of spoken language process-ing, focusing on the question of when during comprehension do idiomatic mean-ings become available.

Even within this perspective, there are many questions one may ask. Oneissue often tackled in this area concerns ambiguity. Many idioms, in fact, areambiguous between a literal and a figurative interpretation. Break the ice, forinstance, can refer either to the crushing of frozen water or to the smoothing of adifficult social interaction. This has led to a concentration of research efforts oninvestigations of which of the two meanings becomes available first to thelistener: the literal, which has to be computed, or the figurative, which is pre-stored in memory and only needs to be retrieved. This line of research, which hasa long tradition (Gibbs, 1980; 1986; Ortony, Schallert, Reynolds, & Antos,1978; Swinney & Cutler, 1979; see also Colombo, chapter 8 of this volume), hasproduced two influential theories of idiom processing.

According to the lexical representation hypothesis, idioms are stored in thelexicon as long, morphologically complex words, and are identified as any otherlexical items. During comprehension, the process of recognition of an idiomstarts at the beginning of the string, in parallel with the computation of its literalmeaning. However, because computing is a longer process than retrieving, theidiomatic meaning of the string becomes available before its literal one (Swinney& Cutler, 1979).

The other view on idiom comprehension assumes that the computation of theliteral meaning of an idiom does not run in parallel with the process of recogni-tion of its figurative meaning. Rather, it occurs after that process, and is startedonly when the idiomatic meaning fails to integrate into context (Gibbs, 1980;1986).'

A somewhat different approach has been taken by Cacciari and Tabossi

'Gibbs recently has modified his views (cf. chapter 3 in this book). His previous positionsnevertheless have been reported here and elsewhere in the chapter because they are very influential inthe area of idiom processing.

7. THE ACTIVATION OF IDIOMATIC MEANING

(1988), who investigated the semantic activation of unambiguous idioms, that is,idioms that do not have a literal meaning (e.g., shoot the breeze). The ex-pressions they used were verbal idioms of two types: predictable and nonpredict-able. Here predictability refers to people's propensity to complete the initialfragment of a string idiomatically. Thus, to be in seventh is completed by mostpeople as to be in seventh heaven, although a literal continuation is possible. Incontrast, to break the tends to be completed literally rather than as to break theice. Accordingly, the former idiom, unlike the latter, is considered to be predict-able.

In three cross-modal lexical decision experiments Cacciari and Tabossi (1988)found different patterns of results depending on whether or not the idiomaticstrings were predictable. When the idioms were predictable, subjects were fasterat performing a lexical decision task to a target presented at the offset of thestring when the target was figuratively rather than literally related to the idiom.However, when the idiom was not predictable, subjects were faster on the liter-ally related target, whereas the idiomatically related target did not differ from thecontrol. Only sometime after the offset of the string, the idiomatically as well asliterally related target were responded to significantly faster than was the control.

In order to explain these findings, and contrary to the view that idioms arerepresented mentally as separate lexical entries (Estill & Kemper, 1982; Glass,1983; Mueller & Gibbs, 1987; Swinney & Cutler, 1979)-henceforth, the lexicalview, the authors proposed that the meanings of idioms, at least verbal ones, areassociated with configurations of words. These words are the same that areprocessed during literal language comprehension, and each takes part in all theidiomatic configurations in which it occurs. According to this hypothesis, theprocessing of an idiomatic string takes place literally, until sufficient informationin the string renders it recognizable as an idiom. Only at this point, referred to asthe idiom key, is the idiomatic meaning activated. Thus, this view is in contrastwith what usually is assumed by current models of lexical processing, accordingto which the activation of the meaning of a word occurs soon after its onset alongwith the activation of the meanings of other words sharing the same onset(Marslen-Wilson, 1987; McClelland & Elman, 1986). The notion of key, whichis discussed further at the end of the chapter, intuitively reflects the point in astring at which one becomes aware of the possibility of being presented with anidiom, and is specified operationally in Cacciari and Tabossi (1988) by means ofthe following procedure: People are requested to complete an increasingly longeridiom fragment, and the point at which the majority of them complete thefragment idiomatically can be considered its key. This point, according to theconfiguration hypothesis, is predictive of when the meaning of the idiom be-comes active.

Admittedly, various factors may speed up the activation of idiomatic mean-ings. A sentential context, for example, can direct one's comprehension of afragment toward its idiomatic completion, rendering the activation of its meaning

147


faster (Tabossi & Cacciari, 1988). Likewise, pragmatic incongruency may pro-vide important cues to idiom comprehension. The occurrence of kicked in thecontext of an old and sick man, for instance, renders the literal action of kickingrather more implausible than the figurative kicking of the bucket, possibly speed-ing up the activation of this meaning. Also, it is well known that people may bebiased toward idioms when presented with them in the context of other similarexpressions (Bobrow & Bell, 1973).

But although linguistic and extralinguistic factors may facilitate the semanticactivation of an idiomatic string, none of these cues is essential to it. In fact, onecommon observation is that idioms are easy to understand; and indeed, manyambiguous idioms appear to be both more commonly used and more quicklyunderstood in their figurative rather than in their literal interpretation (Cronk,1990; Gibbs, 1986). Thus, one tenet of this hypothesis is that at least in theabsence of contextual biases, the semantic activation of an idiom does not takeplace until after its key, whose position in the string determines therefore whenduring the processing of the idiom its meaning becomes available.

Although consistent with the findings in Cacciari and Tabossi (1988), thisassumption was not tested directly in that study. Accordingly, the main goal ofthis chapter is to explore further the notion of key, attempting to provide somedirect evidence for the role that it may play on the activation of idiomaticmeaning during language processing.

EXPERIMENTAL EVIDENCE: EARLYVERSUS LATE: KEY POSITION

The idioms selected for the first cross-modal lexical decision experiment were allwell-known Italian expressions with two main characteristics. First, they wereverbal idioms with two content words after the verb; second, the first contentword was sufficient to prompt the idiomatic completion of the strings. In otherwords, the first content word was the key of these expressions.

The selected idioms, each embedded in a sentence, were presented auditorilyto the subjects paired with a visual target that was related semantically to theirmeaning, as illustrated in the following example:2

Finally, Silvio had succeeded in setting his mind at restRESIGNED

2Experiments I and 2 were conducted with Italian subjects and Italian materials, which are listedin the Appendix along with their corresponding English translations. Because most of the Italianidioms have no exact English equivalent, in the text the examples are English idioms with similarcharacteristics to those employed in the study.


The subjects listened to the sentences and performed a lexical decision task onthe visual targets paired with each idiom. The targets were presented at one ofthree positions: after the verb (e.g., setting), after the first content word (e.g.,mind), and after the second content word (e.g., rest). Differences in the subjects'reaction times (RTs) to decide that the targets were words in the three positionswere taken to reflect the state of activation of the idiomatic meanings at thosepoints.

It was predicted that if the position of the key influences the activation ofmeaning of an idiom during comprehension, here, where the key is the firstcontent word, idiomatic meaning should be already available soon after it.Hence, RTs to the targets at the first and second content word should not differfrom each other and both be faster than at the verb.

EXPERIMENT 1

Method

Subjects. Thirty undergraduates volunteered to participate in the experi-ment, which lasted about 35 min. None had participated previously in an experi-ment of this type.

Materials and Design. Twelve verbal idioms containing two content wordsfollowing the verb were selected. Their syntactic structure was V (NP) (PP),3

where at least one of the components in parentheses was present.These idioms were all well-known expressions in the language. This was

established by asking 15 subjects to paraphrase a list of idiomatic expressions,including the ones used here. These were paraphrased correctly by at least 95%of the subjects.

In order to establish the position of the key in the idioms, 15 independentjudges were asked to complete increasingly longer fragments of a list of idi-omatic strings. In the case of the selected expressions, 90% of the judges com-pleted the string idiomatically after the presentation of the first content word thatwas assumed to be the key of the idioms in this experiment.

Each idiom was embedded in a sentence that was constructed to provide aslittle contextual information as possible. In addition, for each idiom, a wordrelated to its meaning was selected for use as visual target in the experiment (seeAppendix).

Besides the test materials, filler materials also were prepared. They consistedof 96 "formal" and 12 "colloquial" sentences. The latter, which were obtained

3V means verb; NP means noun phrase; PP means prepositional phrase.

149


by using colloquial lexical items but did not contain idiomatic expressions, weredevised in order to avoid the test sentences being noticed by the subjects asstrange or awkward in the experimental situation. Each of the 108 filler sen-tences, which were approximately of the same structure and length as the testsentences, was paired with a visual target that had no relation with either theoverall meaning of the sentences or any of their individual words.

Out of the 96 formal sentences, hall were paired with target visual words,whereas the remaining 48 formal sentences in addition with the 12 colloquialones were paired with targets that were legal pseudowords.

All 120 sentences (12 test and 108 filler) were recorded by a male speaker onone channel of a tape recorder. There was a 5-s interval between sentences. A1,000 Hz pulse, placed automatically on the other channel of the tape andinaudible to the subjects, caused a word to be displayed on the screen of an Applemicrocomputer for 1,500 ms. It also started a digital timer, which either stoppedwhen the subjects pressed the space bar or reset automatically after 5 s.

In the filler sentences the pulse occurred 100 ms after the offset of one wordchosen in such a way as to cover the entire range of positions within sentences(i.e., beginning, middle, end).

All the experimental sentences were recorded three times, so that for each ofthem the pulse could be placed in three positions: 100 ms after either the verb,the first content word, or the second content word.

Pulses were placed 100 ms after word offsets rather than at their exact end inorder to maximize the probability of :he idiomatic meaning being detected ateach position, because there is reason to believe that the activation of an idiommay be a slow process, and its meaning may require sometime after string offsetbefore becoming detectable (Cacciari & Tabossi, 1988).

In order for the same idiom to be presented only once to each subject, threelists of sentences were constructed, each containing only one version of each testsentence. Each list contained all 12 idioms: 4 with the pulse after the verb, 4 afterthe first content word, and 4 after the second content word. In addition to the testsentences, each list also included the 1()8 filler sentences, thus yielding a total of120 sentences. One set of 120 visual targets (12 idiomatic, 48 filler word, and 60nonword targets) was paired with the three lists of sentences.

Each list of sentences, together with the paired set of targets, was divided inthree blocks of 40 trials each. Trials were counterbalanced across blocks, andoccurred in a quasi-random order, with the constraint that no more than threetarget words or nonwords appeared in a row.

An equal number of subjects was assigned randomly to one of the experimen-tal lists. Within each list, subjects acted as their own controls.

Procedure. Subjects were tested individually. They sat in a sound-attenu-ated room in front of an Apple microcomputer monitor connected to a stereo taperecorder. They were instructed to perform a lexical decision task to strings of


letters that would appear on the monitor while listening to a sentence. They weretold to press the space bar of the keyboard with their dominant hand if the visualtarget was a real word and to do nothing otherwise. After 10 practice trials, oneof the three experimental lists was presented to each subject. Reaction times tolexical decisions were recorded.

The subjects were informed that they should pay attention to the sentencesbecause at the end of the test session they would be asked questions about them.Immediately after the end of the experimental session, the subjects were given 20sentences, each printed on a card, half of which had been presented in theexperiment. The other half was derived from the sentences the subjects actuallyheard. The subjects' task was to decide which sentences were old and which werenew. The mean percentage of correct recognitions was 73%, and none of thesubjects recognized less than 60% of the sentences.

Results and Discussion

In order to reduce variability, data points above or below 2 standard deviations(SDs) from the mean RTs to each target were excluded from the analyses. Thesedata, together with errors, accounted for 3.33% of all responses. Table 7.1 showsthe mean RTs and SDs for the correct responses under the three experimentalconditions.

There was a reliable main effect both in the analysis by subjects [F(2, 58)5.34, MSe 3,174; p = .0071 and in the analysis by materials [F(2, 22) = 4.77,MSe 1,670; p = .02]. Planned nonorthogonal comparisons showed that RTs tothe targets were faster after the first content word than after the verb, whereasRTs after the first and the second content word did not differ significantly fromeach other: V versus Wl, F(l, 58) = 5.64, p = < .05; Wl versus W2: F(l, 58)= 0.70, n.s.).

In agreement with the predictions, the findings suggest that in those idioms inwhich the key is the first content word, meaning is already active after that word.Unfortunately, this pattern of results is also compatible with the view that idioms

TABLE 7.1Mean RTs and SDs for Correct Lexical Decisions

to Targets in the Three Positions: V (Verb),W1 (First Content Word), and W2 (Second Content Word)

in Experiments 1 and 2

V W1 W2

Experiment 1 672 638 626(96) (78) (74)

Experiment 2 637 631 605(69) (81) (78)

151


are represented mentally as lexical items. If idioms are represented in the sameway as words, in fact, they also should be processed in a comparable way, andhence have their meanings activated early during identification, in analogy withwhat happens in lexical processing.

The configuration and the lexical alternatives lead, however, to differentpredictions in the case of idioms that, unlike those in Experiment 1, have theirkey later than the first content word, toward the end of the string. According tothe lexical view, the time course of activation of the meanings of these ex-pressions should not be different from that of the idioms in Experiment 1. In bothcases, meaning should be activated very early on in the string, in analogy withwhat happens in lexical processing. A ccording to the configuration hypothesis,however, the further down in the string an idiom key is, the later will its meaningbe activated. Hence for idioms whose key occurs later than in the idioms inExperiment 1, activation also should be found later, and not be already availableat the first content word. The prediction was tested in a further experiment. Thiswas analogous to the previous one, except that idioms used here had their key atthe second content word rather than at first, as in the following example:

In the end, the man hit the nail on the head.ACC RATE

The predictions were as follows. If the activation of the meaning of an idiomdoes not occur until its key has been processed, in this experiment no facilitationto the targets should be observed before the second content word. Hence, RTs tothe targets should be faster in this position than in the other two positions, whichshould not differ from each other. Alternatively, the notion of key does not play arole in the meaning activation of an idiom. In this case, idiomatic meaningactivation should be initiated already after the first content word, and RTs to thetarget at this point therefore should be significantly faster than after the verb,where it is presumably too early for idiomatic meaning to be active and prime thetarget.

EXPERIMENT 2

Method

Subjects. A total of 27 undergraduates volunteered for the experiment,which lasted about 35 min. As in Experiment 1, none had participated previouslyin an experiment of this sort.

Materials and Design. A new group of 12 familiar idioms was chosen.These idioms had the same syntactic structure as those in Experiment 1. Also


they had two content words following the verb. Their key, however, occurredlater than in the previous group. This was established on the basis of the samepretest reported in Experiment 1, but the criterion for selection differed, in thatnone of the new idioms was completed idiomatically after the first content word.In fact, 90% of the judges failed to complete idiomatically the whole strings priorto the second content word. Thus this word was taken to be the key of theseexpressions (see Appendix). The filler materials were the same as those used inExperiment 1, and so was the design.

Procedure. The procedure was the same as in the Experiment 1. In thememory task, the mean of correct recognitions was 75%, and none of the sub-jects had a performance below 60% correct identifications.

Results

The mean of errors was 2.16%. The mean RTs of correct responses in the threeexperimental conditions and related SDs are reported in Table 7.1. Analyses ofvariance, both by subjects and by materials, showed a reliable main effect oftarget position: F1(2, 52) = 3.71, MSe 2,155;p = .03; F2(2, 22) = 3.34, MSe1,161; p = .054.

Planned nonorthogonal comparisons showed no significant difference be-tween RTs after the verb and after the first content word, which were bothreliably slower than RTs after the second content word: V versus WI, F( 1, 52) =0.26, n.s.; WI versus W2, F(1, 52) = 4.23, p = .04.

The data indicate that the meaning of an idiom is not activated until very lateduring the processing of that expression, in agreement with the prediction of theconfiguration hypothesis for those expressions whose key, as in this experiment,is toward the end of the string.

ACTIVATION AND KEY

Taken together, Experiments I and 2 suggest that the notion of key does play arole in determining idiomatic meaning activation. In fact, the point of activationof an idiom during its processing depends on the characteristics of the string:When its key occurs early, its meaning is also activated early (Experiment 1);when the key is later, so too is activation (Experiment 2).

These findings also support the hypothesis that the meaning of an idiom is notactivated soon after its onset; rather it is triggered only after sufficient informa-tion is available to the listener to signal the presence of an idiom. Only after sucha point-that is, after the key-does the idiomatic meaning become available.

It might be argued that this interpretation of the results is unwarranted. In fact,although indicating differences in the semantic activation of the two groups of

153


idioms at the first and at the second content word, the data fail to show that thereis no activation for either group at the verb. This point is crucial. Let us suppose,in fact, that such activation exists, and RTs after the verb are not a baseline, butreflect the initial semantic activation of idiomatic meaning. In this case, findingsin Experiments I and 2 simply would indicate differences in the speed at whichsemantic activation builds up in the two groups of idioms, actually supporting thelexical view that predicts early activation for all idiomatic expressions.

Two arguments run against this interpretation. First, the differences betweenverb and first and second content word in Experiment 1 (34 ms and 46 ms,respectively), and between verb and second content word in Experiment 2 (32ms) are typical of the range of differences observed between primed and un-primed conditions in cross-modal lexical decision studies (Swinney, 1979; Ta-bossi, 1988; Zwitserlood, 1989). Also, evidence to the effect that idiom mean-ings are not yet active at the verb comes from Swinney (1982), who reported across-modal study on the time course of activation of expressions such as see thelight, where he found the idiomatic meaning active at the end of the string, butnot at verb offset.

The present findings are therefore difficult to reconcile with the view thatidioms are processed as long words. In the field of lexical processing, it gener-ally is agreed that semantic information is activated before the identification of aword (Marslen-Wilson, 1987). As soon as a word fragment is available, themeanings of various words compatible with the perceptual input up to that pointare activated, and subsequent perceptual information is used to select among theactive competing meanings the one associated to the actual stimulus word.

Many aspects of these processes are still under debate, including the questionof what is the perceptual information used to initiate word recognition, or whatare the words whose meanings are activated during early phases of the process(Bard & Shillcock, in press; Lahiri & Marslen-Wilson, 1991; Luce, 1986; Note-boom, 1981; Salasoo & Pisoni, 1985). However, it commonly is accepted thatmore than one meaning is activated after the initial fragment of a spoken wordand this process of multiple activation occurs early, before word recognition.Hence, if the lexical view of idiom representation and processing is correct,halfway through the expressions one ought to find their meaning always active,and no difference should be observed between idioms in Experiments I and 2.

Perhaps one way of reconciling the present data with a lexical hypothesis ofidiom processing would be to explain the pattern of results in this study as due tothe different uniqueness or recognition points of the idioms in Experiments I and2. In lexical processing, uniqueness point refers to the point at which on the basisof perceptual information a word diverges from any other in the language,whereas recognition point indicates the point at which a word can be discrimi-nated from the other competing candidates, taking into account contextual aswell as sensory constraints. Thus, it might be claimed that the notion of key of anidiom corresponds to the uniqueness point of a lexical item, or perhaps to its


recognition point, and has in idiom processing the same role that uniqueness andrecognition point have in word processing. Here, it is well known that the earlieris the uniqueness point in a lexical item, the sooner this is recognized. Moreover,recognition may be anticipated further in an appropriate context: There the selec-tion of a word meaning for contextual integration may take place before itsuniqueness point. Analogously, idiom meanings were activated sooner in Experi-ment I than in Experiment 2 simply because the key, which according to thisargument corresponds to their uniqueness or recognition point, occurred earlierin the former group of expressions than in the latter.

It should be noted, however, that although similar to both the uniqueness pointand the recognition point of a word, the key of an idiom cannot be characterizedin the same way as either of these notions. In fact, the uniqueness point of a wordin a language is determined by the lexicon of that language. Although pow- canbe completed as powder, power, or pow-wow, powd- can be completed only aspowder: /d/ is accordingly the uniqueness point for that word because no otherwords in English have powd- as their initial fragment. This is not the case withthe key of an idiom, which may occur at a point where the perceptual input stillcould have a literal completion. To be in seventh, for instance, could be com-pleted equally well with position or heaven. In fact, most of the predictableidioms in Cacciari and Tabossi (1988) were of this sort.

Likewise, although the recognition point of a word can precede its uniquenesspoint, because contextual factors intervene along with perceptual factors to deter-mine word identification, the key of an idiom is not established contextually, andthere is some evidence that predictable strings are completed idiomatically evenin contexts that do not rule out literal completions. "The tennis player was inseventh . . .," for instance, is usually completed as "The tennis player was inseventh heaven," even though "The tennis player was in seventh position" is anequally good completion (Cacciari & Tabossi, 1988). Thus, the key of an idiomseems to correspond exactly neither to the uniqueness nor to the recognition pointof a lexical item, and it is therefore unlikely that its role in idiom processing canbe reduced simply to the role of either notion in word recognition. Furthermore,even if the key of an idiom did correspond to word uniqueness or recognitionpoint, these notions still would not explain the pattern of results. Both recogni-tion and uniqueness point, in fact, refer to the point in time where, in isolation orin context, a word diverges from all its alternatives. According to the multipleactivation hypothesis of lexical processing, however, the meaning of a word isactivated, along with other lexical meanings, before either point. Yet, in contrastto what seems to be the case with lexical processing, the present findings indicatethat semantic activation of an idiom occurs late in the string, and how late maydepend on factors such as the location of the key, for which there is no obviouscorrespondence in the processing of individual lexical items.

But how can a key be characterized? Although this notion, as noted earlier,can be operationalized easily, its theoretical specification is less simple. The

155


difficulty-or at least part of it-stems from the fact that as already pointed out,a key does not seem to have any formal property: It does not correspond to apoint after which the language offers no other acceptable possibility than com-plete the string idiomatically; it has no special syntactic role, nor does it neces-sarily mark a syntactic or semantic anomaly.

Perhaps the best way of specifying an idiom key is in probabilistic terms, asthe point in the string after which the probability of the fragment to continueidiomatically is very high, even though a different, literal completion still may bepossible. Probability here refers to language use. Thus, although both buildcastles and hit nails can refer to perfectly reasonable actions, in actual use thefirst expression tends to be employed in the context of building castles in the air,whereas the latter often is used with reference to hammers, picture frames, andthe like. There is no intrinsic necessary difference between those idioms, but inmany varieties of contemporary English the former fragment is far more predic-tive of the occurrence of an idiom than the latter, and people seem to be sensitiveto this factor while processing those expressions.

Admittedly, this characterization is at best incomplete. It fails to capture thefact, for instance, than an idiom key not only indicates a point in the string, butoften is the most important part of the idiom. A common observation, in fact, isthat not all elements in an idiomatic expression seem to contribute equally to itscomprehension, some parts being more crucial than others. This is illustrated bythe fact that many idioms (e.g., break ihe ice) retain their meaning even withsome lexical changes (e.g., crack the ice), but there are words that cannot besubstituted without loss of meaning (e.g., break the frozen water).

But although partial, probability does seem to have a role in the characteriza-tion of an idiom key. In fact, it has some similarities with the transitionalprobability used by behaviorists to describe linguistic behavior, when this wastreated as a stochastic process (Shannon & Weaver, 1949). Within that frame-work, the transitional probability of a word is the probability of occurrence ofthat word, given the preceding words. Clearly, at the beginning of a sentence,almost if not all words can occur, but as more and more information comes in,the possible continuations offered by the language, and by discourse and knowl-edge constraints become increasingly limited, up to a point where only onepossibility is left (Attneave, 1959). The critiques to this approach are not onlywell known, but are part of the foundations of cognitive psycholinguistics(Chomsky, 1957, 1959). However, the notion of transitional probability seems tocapture one characteristic of an idiom key: Namely, it does not reflect any formalproperty of idioms, but the fact that when certain words co-occur in the languagevery frequently they are part of an idiom.

The importance of a factor such as use on idiom processing also is suggestedby a number of studies that have looked at their frequency and familiarity(Cronk, 1990; Popiel & McRae, 1988; Schweigert, 1991). Popiel and McRae,for instance, had their subjects rating how familiar they were with a list of


ambiguous idioms, and how frequently those expressions occurred in their literaland figurative meaning. The results indicated that among familiar idioms, someoften are used both literally and figuratively, but others occur almost exclusivelyin their figurative meaning.

That frequency factors may play a role in how early idiomatic meanings areactivated during language processing is not surprising, given the well-establishedand robust effects of frequency in almost all domains of language comprehen-sion. What is specific of the way in which frequency-and hence probability-isconsidered here is that it refers not to a single element, but to the co-occurrenceof items in a sequence of elements, namely the idiom string.

In conclusion, from the perspective of language processing, the empiricalevidence presented in this study appears to support the view that the meaning ofan idiom is not activated as that of a word. Rather, processing an idiomatic stringis probably not very different from processing proverbs, song lyrics, or poems.Just like what happens with idioms, in fact, it is most improbable that whenlistening to "La nebbia . . ." or "La nebbia agli . . .", anyone who has attendedItalian schools activates anything besides the individual words in the string.However, if the next word is irti, a long-known poem suddenly pops up frommemory, and she or he can do nothing but get ready for the worst.4

ACKNOWLEDGMENTS

This research was supported by Fondi 60% Grant No. 9000107 and CNR No.892361. We would like to thank Cristina Cacciari, Corrado Cavallero, SamGlucksberg, and Donia Scott for their helpful comments and suggestions.

REFERENCES

Attneave, F. (1959). Applications of information theory to psychology. New York: Holt, Rinehart &Winston.

Bard, E. G., & Shillcock, R. C. (in press). Competitor effects during lexical access. In G. T.Altmann & R. C. Shillcock (Eds.), Spoken language processing. Hillsdale, NJ: LawrenceErlbaum Associates.

Bobrow, S., & Bell, S. (1973). On catching on to idiomatic expressions. Memory & Cognition, 1,343-346.



Chomsky, N. (1957). Syntactic structures. The Hague, Netherlands & Paris: Mouton.Chomsky, N. (1959). A review of B. F. Skinner's "verbal behavior." Language, 35, 36-58.

4 The unforgettable poem is San Martino by Giosue' Carducci.

157


Cronk, B. (1990). Familiarity and likeliness fir figurative and literal meanings of idioms. Posterpresented at the 2nd annual convention of 1he American Psychological Association, Dallas.

Dascal, M. (1987). Defending literal meaning Cognitive Science. 11, 259-281.Estill, R. B., & Kemper, S. (1982). Interpreting idioms. Journal of Psycholinguistic Research, 11,

559-568.Fernando, C. (t978). Towards a definition of idiom, its nature and function. Studies in Language,

2(3), 313-343.Fraser, B. (1970). Idioms within a transformational grammar. Foundations of Language, 6(1), 22-

42.Gibbs, W. R. (1980). Spilling the beans on understanding and memory for idioms in context.

Memory & Cognition, 8, 149-156.Gibbs, W. R. (1986). Skating on thin ice: Literal meaning and understanding idioms in conversa-

tion. Discourse Processes, 9, 17-30.Gibbs, W. R., & Nayak, N. P. (1989). Psychol [nguistic studies on the syntactic behavior of idioms.

Cognitive Psychology, 21, 100-138.Gibbs, W. R., & O'Brien, J. E. (1990). Idioms and mental imagery: The metaphorical motivation

for idiomatic meaning. Cognition, 36, 35-68.Glass, A. L. (1983). The comprehension of idioms. Journal of Psycholinguistic Research, 12, 429-

442.Katz, J. J. (1973). Compositionality, idiomaticity and lexical substitution. In S. R. Anderson & P.

Kiparsky (Eds.), A festschrift for Morris Halle (pp. 357-376). New York: Holt, Rinehart &Winston.

Lahiri, A., & Marslen-Wilson, W. D. (1991). Thie mental representation of lexical form: A phono-logical approach to the recognition lexicon. Cognition, 38, 245-294.

Lakoff, G. (1986). The meanings of literal. Metaphor and Symbolic activity, 1(4), 291-296.Luce, P. (1986). Neighborhoods in the mental lexicon. Unpublished doctoral dissertation, Indiana

University, Bloomington.Marslen-Wilson, W. D. (1987). Functional parallelism in spoken word recognition. In U. Frauen-

felder& L. K. Tyler (Eds.), Spoken word recognition (pp. 7 1 -102). Cambridge, MA: MIT Press.McClelland, J. L., & Elman, J. L. (1986). The TRACE model of speech perception. Cognitive

Psychology, 18, 1-86.Mueller, R. A., & Gibbs, R. W. (1987). Processing idioms with multiple meanings. Journal of

Psycholinguistic Research, 16, 63-81.Noteboom, S. D. (1981). Lexical retrieval from fragments of spoken words: Beginnings versus

endings. Journal of Phonetics, 9, 407-424.Ortony, A., Schallert, D. L., Reynolds, R. E., & Antos, S. J. (1978). Interpreting metaphors and

idioms: Some effects of context on comprehension. Journal of Verbal Learning and VerbalBehavior, 17, 465-477.

Popiel, S. J., & McRae, K. (1988). The figurative and literal uses of idioms, or all idioms are notused equally. Journal of Psycholinguistic Research, 17(6), 475-487.

Rumelhart, D. E. (1979). Some problems with the notion of literal meaning. In A. Ortony (Ed.),Metaphor and thought (pp. 78-90). Nev York: Cambridge University Press.

Salasoo, A., & Pisoni, D. B. (1985). Sources of knowledge in spoken word identification. Journalof Memory and Language, 24, 210- 231.

Schweigert, W. A. (1991). The muddy waters of idiom comprehension. Journal of PsycholinguisticResearch, 20, 305-314.

Shannon, C. E., & Weaver, W. (1949). The mathematical theory of communciation. Urbana: Uni-versity of Illinois Press.

Swinney, D. A. (1979). Lexical access during sentence comprehension: (Re)consideration of con-text effects. Journal of Verbal Learning and Verbal Behavior, 14, 645-660.


Swinney, D. A. (1982). The structure and time course of information interaction during speechcomprehension: Lexical segmentation, access, and interpretation. In J. Mehler, E. C. T. Walker,& M. Garrett (Eds.), Perspectives on Mental Representation (pp. 151-167). Hillsdale, NJ:Lawrence Erlbaum Associates.

Swinney, D. A., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journalof Verbal Learning and Verbal Behavior, 18, 523-534.


Tabossi, P., & Cacciari, C. (1988). Context effects in the comprehension of idioms. In CognitiveScience Society (Eds.), Proceedings of the Tenth Annual Conference of the Cognitive ScienceSociety (pp. 90-96). Hillsdale, NJ: Lawrence Erlbaum Associates.

Wasow, T., Sag, I., & Nunberg, G. (1983). Idioms: An interim report. In S. Hattori & K. Inoue(Eds.), Proceedings of the XlIth International Congress of Linguistics (pp. 102-105). Tokyo:CIPL.

Zwitserlood, P. (1989). The locus of the effects of sentential-semantic context in spoken-wordprocessing. Cognition, 32, 25-64.

APPENDIX

The following lists are the test sentences and targets for Experiments I and 2 inItalian, translated word by word into English and with the meanings of the Italianidioms. In most cases, these do not have an English equivalent, so in order torespect the form of the idiomatic expression in Italian the translation into Englishis necessarily stilted.

Experiment 1

1. Finalmente, Silvio era riuscito a mettersi il cuore in pace.-RASSEGNATOFinally, Silvio had succeeded in setting his heart at rest.-RESIGNED

2. Al lavoro ogni giorno saltava la mosca al naso a qualcuno.-IRRITABILEAt work every day it jumped the fly to the nose of someone. (someonewas upset)-IRRITATED

3. Giovanna, che era una nuova impiegata, prendeva a pesci in faccia tutti.-MALEJoan, who was a new clerk, took with fishes in the face everybody. (ill-treated everybody).-BAD

4. Alla fine, si scopri che l'opera dello scrittore non era farina del suo sacco.-TRUFFA

159


At the end, it was discovered that the work of the writer was not flour ofhis bag. (his own work)-CHEAT

5. Improvvisamente, ii senatore americano pass dalle stelle alle stalle.-PEGGIOSuddenly, the American senator went from stars to stables. (lost impor-tance)-WORSE

6. Per alcuni e naturale lavare un offesa col sangue.-VENDETTAFor someone it is natural to wash an offence with blood. (kill the offen-ders)-REVENGE

7. Come sempre, Cristina tirava acqua al suo mulino.-VANTAGGIOAs usual, Christine pulled water to her mill. (pursued her interest)-ADVANTAGE

8. Dopo ii colloquio col direttore, Luca aveva un diavolo per capello.-RABBIAAfter the meeting with his boss, Luc had a devil each hair. (was furious)-FURY

9. Tutti sapevano che la ragazza diceva pane al pane.-SINCERAAll knew that the girl used to say bread to the bread. (talked the truth)-SINCERE

10. Secondo l'opinione degli amici, Giovanni era in una botte di ferro.-SICUROAccording to the opinion of his friends, John was in a barrel of iron. (wason the safe side)-SAFE

11. Claudio toccava il cielo con urn dito ogni volta che usciva con Milena.-FELICEClaude touched the sky with a finger (was happy) everytime he went outwith Milena.-HAPPY

12. La fidanzata pensava che a Marco desse di volta il cervello.-PAZZOThe girlfriend thought that to Marc the brain turned. (was mad)-MAD


Experiment 2

1. Nel romanzo, il personaggio principal faceva l'avvocato del diavolo.-POLEMICOIn the novel, the main character was the devil's advocate.-POLEMIC

2. L'anziana signora aveva il cuore in gola quando vide entrare glisconosciuti.-PAURAThe old lady had the heart in the throat (was frightened), when she sawstrangers come in.-FEAR

3. La madre si accorse che il figlio aveva le mani bucate.-SOLDIThe mother noticed that the son had the hand pierced. (was a spendthrift)-MONEY

4. Alla cena, Maria vide con stupore che c'era un pesce fuor d'acqua.-IMBARAZZOAt the dinner party, Mary saw with surprise that there was a fish out ofwater.-EMBARRASSMENT

5. Lo scolaro mise una mano sul fuoco che il suo compagno non mentiva.-SICUROThe boy put a hand on the fire (swore) that his schoolmate did not lie.-SURE

6. 11 bambino sapeva benissimo quale era il cavallo di battaglia della suacantante favorita.-SUCCESSOThe child knew very well which was the warhorse of his favorite singer.-SUCCESS

7. Nel film, il protagonist aveva una spada nel cuore.-DOLOREIn the movie, the main character had a sword in the heart. (pain)-PAIN

8. Sandro si ricordava ancora di quando vedeva il sole a scacchi.-PRIGIONESandy still remembered when he used to see the sun in checks. (was injail)-JAIL

161


9. Stefano sapeva che finchM non era verde per la bile, il padre taceva.-ARRABBIATOSteve knew that until he was green for the bile (was furious), the fatherkept silent.-ANGRY

10. L'insegnante mostr6 quale era la punta di diamante della sua classe.-MIGLIOREThe teacher showed which was the diamond point in her class.-BEST

11. Secondo la signora, era la pietra dello scandalo della famiglia la causa ditutto.-CATTIVOAccording to the woman, it was the stone of the scandal (the culprit) ofthe family the cause of everything.-BAD

12. La sola cosa spiacevole della vacanza fu la spina nel fiance delle dif-ficolta economiche.-PROBLEMAThe only unpleasant thing in the vacation was the thorn in the side(aching pain) of the financial problems.-DIFFICULTY

8 The Comprehension ofAmbiguous Idioms in Context

Lucia ColomboUniversity of Padova, Italy

The study of the identification and comprehension of ambiguous idiomaticstrings, like kick the bucket, shares many of the issues that are involved in thestudy of lexical ambiguity, and as such is important for understanding the pro-cesses involved in sentence comprehension. Ambiguous idioms in fact can beassigned either a literal interpretation, based on the compositional meanings ofits constituents, or a figurative meaning, which has been learned and associatedwith the whole string. Some of the crucial questions that have been asked in thestudy both of ambiguous words, and of ambiguous idioms, concern problemslike the following: Is there an automatic exhaustive activation of meanings or acontext-dependent preactivation? At which point in time does activation of eachmeaning occur? Is selection of the appropriate meaning immediate or delayed?

Because of their relevance to the study of ambiguous idiomatic expressions,the main models of lexical ambiguity are described briefly. The two main viewsthat have been proposed to describe the temporal course of activation of mean-ings in ambiguous words are the exhaustive access model and the selectiveaccess model. According to the first, all meanings are activated automaticallyindependent of context bias and of meaning dominance, and selection of theappropriate meaning is made in a later stage (Onifer & Swinney, 1981; Seiden-berg, Tanenhaus, Leiman, & Bienkowski, 1982; Swinney, 1979; Tanenhaus,Leiman, & Seidenberg, 1979). According to the selective access model, contextcan preactivate one meaning (Simpson, 1981; Tabossi, 1988; Tabossi, Colombo,& Job, 1987; Van Petten & Kutas, 1987). When one of the meanings is dominantwith respect to the others, because of its frequency of occurrence, it will beactive earlier independently of context, whereas the secondary meanings needpreactivation by contextual features to become active. A variation of this model,

163

164 COLOMBO

the reordered access model, was proposed by Duffy, Morris, and Rayner (1988)and Rayner and Duffy (1986). According to the latter model, the effect of contextand dominance is to make the biased meaning available earlier, rather than toselectively preactivate it.

Although the selective access and the reordered access models localize theeffect of context before or during semantic activation, another model was pro-posed by Rayner and Frazier (1989: see also Frazier & Rayner, 1990), theintegration model, in which it is claimed that all meanings are activated automat-ically independent of context, and the effect of context is located during theprocess of integration. Ease of integration, together with dominance of meaning,account for the rapidity with which a meaning is selected and the nonsuitablemeaning is discarded. The interesting feature of this model is that it allows theassumption of an automatic activation of meanings, and at the same time itaccounts for the effect of dominance and context.

The roles of context and of the availability of meanings during the temporalcourse of activation and integration are also important to the study of the com-prehension of ambiguous idioms. This becomes apparent when examining themodels that have been proposed to explain the comprehension of idiomaticexpressions.

Quite opposing views are represented in the two models called the idiom listhypothesis and the direct access hypothesis. According to the idiom list hypoth-esis (Bobrow & Bell, 1973), idioms are represented in a special list in the lexiconand are retrieved from this list only after the literal interpretation has failed tomatch the context.

More recently, Gibbs and collaborators (Gibbs, 1986; Gibbs & Gonzales,1985) conducted a series of experiments showing that idioms can be accesseddirectly, and the literal interpretations need not be computed. According toGibbs, idioms are stored in the normal lexicon, and the ease with which they areaccessed (as single words) depends on their degree of "frozenness." This termhas been used to indicate the extent to which idiomatic expressions can undergosyntactic transformations without changing the figurative meaning (Fraser,1970). When the words of the idiom are connected strongly, they are more likelyto be treated as single units. Once the idiomatic meaning becomes active, theliteral interpretation that was being constructed is inhibited (Gibbs, 1980, 1986).Gibbs suggested that the context in which the idiomatic expression is encoun-tered can be used to activate the idiom directly.

This suggestion is in contrast both with the idiom list hypothesis, whichclaims that the literal interpretation is always the first to be available, and withthe view held by Swinney and Cutler (1979), the lexical representation hypoth-esis. According to the latter, idioms are represented as large words, and as suchthey are activated as soon as enough perceptual information has been extracted.No matter how biasing is the context, according to this model both the idiomaticand the literal meanings of the phrase are activated simultaneously. The lexical

8. COMPREHENSION OF AMBIGUOUS IDIOMS IN CONTEXT

representation hypothesis is analogous to that proposed in the exhaustive accessmodel. The underlying idea is that lexical access is an autonomous process,which takes place automatically, independently of other higher level processesand before selection of the meaning appropriate to the context occurs.

Applied to the processing of idioms, the exhaustive access model predicts thatwhen an idiom is encountered, both its interpretations are accessed simul-taneously. However, ambiguous idioms are not completely comparable to ambig-uous words because the retrieval of meanings in the case of ambiguous wordsinclude operations at a level not higher than lexical access. In the case of idioms,retrieval of the idiomatic meaning indeed may be limited to lexical access, if theassumption holds that idioms are represented as single words. However, thecomputation of the literal meaning involves both lexical access of the singlewords, and a structural and interpretative analysis at the sentence level. Per-forming these operations may require different time courses, as it appears, forinstance, from the experiments reported in the literature (Gibbs, 1980; Glass,1983; Ortony, Schallert, Reynolds, & Antos, 1978; Swinney & Cutler, 1979),which show that comprehension of the figurative meaning of the idiom is fasterthan comprehension of the literal meaning. For instance, deciding whether asentence is meaningful or not is faster for idiomatic expressions, as compared tonormal, literally interpreted phrases (Swinney & Cutler, 1979), and is faster formore frozen (that is, less suitable to transformations without a change in mean-ing) than less frozen idioms (Gibbs & Gonzales, 1985).

The reason why different time courses probably are implicated might lieprecisely in the fact that if idioms are represented as single words, the meaningassociated with each of them is already stored, or precomputed, whereas a literalinterpretation must be constructed through inferences or the computation ofrelations among the single words. One possibly could claim that at the beginningof the idiomatic phrase only the literal interpretation has started. By the time thelast word begins, however, enough information may have accumulated to allowactivation of the idiom, and access to its stored meaning. By this time, process-ing of the literal interpretation already may have stopped and decayed. Thisinterpretation has received support by experiments using the cross modal primingparadigm (Cacciari & Tabossi, 1988; Swinney, 1981).

In Swinney's (1981) experiment, neutral context sentences containing an idi-omatic expression (e.g., kick the bucket) were presented auditorily; target wordswere presented visually, either at the offset of the first word of the idiom or at theoffset of the last word. He found that at the end of the first word (kick) only theliteral meaning was activated, whereas at the end of the idiom only the targetrelated to the figurative meaning was facilitated.

Swinney's (1981) results are inconsistent with the lexical representation hy-pothesis in two ways: first, because they show that the figurative meaning is notactivated automatically from the beginning of the idiom, as suggested by thelexical representation hypothesis. More important, the literal meaning does not

165

166 COLOMBO

appear to be activated at the end of 1he idiom, in contradiction with the viewholding a simultaneous activation of literal and figurative meanings. Swinney'sresults also suggest that a triggering context is not necessary for the activation ofthe idiomatic meaning to take place, because it appears to be activated even in aneutral context.

A similar result was found by Cacciari and Tabossi (1988) with predictableidioms that were unambiguous, that is, idioms that cannot be understood literallybut that can be interpreted literally before the last word is encountered (e.g.,shoot the breeze). However, when unpredictable idioms were used there was noevidence of activation of the figurative meaning when the last word had just beenheard. Only after 300 ms from the offset of the idiom was there evidence offacilitation for reaction times to the target related to the figurative meaning. Inthese experiments the context was neutral. When a biasing context was provided(Tabossi & Cacciari, 1988), both the literal meaning of the last word of the idiomand the figurative meaning were primed at the offset of the final word of theidiom. These results are particularly striking because the idioms used by Cacciariand Tabossi did not have a literal interpretation after the last word and suggestthat at least the literal meaning of the single words, if not the structural literalinterpretation, always is activated automatically. It is another matter, of course,to decide whether the literal interpretation of the whole string is initiated even incontexts biasing the figurative interpretation of an idiom, or when idioms with noliteral interpretation are processed.

One of the main characteristics of idioms, which has been considered criticalfor their definition (Chomsky, 1980; Fraser, 1970), is that the figurative meaningcannot be derived from a compositional interpretation of the literal meanings ofthe single words that compose the idiom. If it is assumed that literal meanings areactivated automatically and that the language processor proceeds with the con-struction of a structural interpretation and an elaboration based on compositionalmeanings, then whether a figurative meaning is activated and the moment inwhich it is activated may depend on several factors. Thus, it is crucial to deter-mine what factors are responsible for triggering the activation of its figurativemeaning.

The aim of the present chapter is to investigate the effect of context on theprocessing of ambiguous idioms. In the first experiment a cross-modal primingparadigm was used. A sentence containing an idiom and biasing either its literalor its figurative interpretation was presented auditorily; at the offset of the lastword of the idiomatic phrase a target word was presented visually, related eitherto the literal meaning of the last word of the idiom (kick the bucket, pail) or to itsfigurative interpretation (kick the bucket, die). When either the literal or thefigurative meaning of the idiomatic expression has been retrieved, lexical deci-sion on a target word related to the accessed meaning should be facilitated, withrespect to the lexical decision made on a control word. If facilitation with respectto the control word is obtained from target words related to both meanings, then


it may be concluded that both the literal meaning of the last word of the idiomand the figurative meaning have been retrieved simultaneously.

In the present experiment two groups of subjects were used. One group waspresented with the context that biased only the literal reading of the idioms, thesecond group with the context biasing the figurative meaning. Subjects whoalways were presented with idioms in the literal reading should be less likely toshow an effect of expectations about figurative expressions. Thus, priming of theidiomatic meaning should not appear in the literal bias context, and if it does,then it would be even stronger evidence that the figurative meaning is activatedautomatically. If, on the other hand, an effect of context occurs, it should bemanifested as a differential facilitation for targets related to either one of thereadings, as a function of context.

EXPERIMENT 1

Method

Subjects. Two groups of 20 subjects each were tested, all of them studentsof psychology at the University of Padua.

Material. In the first phase of the selection 63 idiomatic phrases and 64nonidiomatic phrases of about the same length were presented to 31 subjects,with the task of writing down the first word that came to their mind after readingthe phrase. The aim of the task was to select ambiguous items, that is, items withthe literal meaning as frequent as the figurative. This was ascertained by check-ing whether the associated word was related to either of the two meanings. Forinstance, if the words associated to the idiomatic phrase kick the bucket were pailor water, they were considered to be related to the literal meaning. If the associ-ated words were die or related words, then the meaning implied was the figur-ative one. On the basis of the association task, and of the author's judgment, 16items were selected that had equally plausible literal and figurative meanings.The items were of about the same length in words.

Two words were paired to each idiom, related to either its literal or itsfigurative meaning, to be used as target words in the lexical decision task. Thetarget relative to the literal sense was associated or related semantically to the lastword of the idiom, whereas the target relative to the figurative meaning wasrelated to the meaning of the whole expression. In addition, two control targetwords were selected, matched for length and frequency to the respective relatedtarget. The mean frequency of the target words was 44, 19, 38, and 20 (on acount of 500,000; Bortolini, Tagliavini, & Zampolli, 1972) for, respectively,targets related to the literal meaning, targets related to the figurative meaning,and the respective controls.

167

168 COLOMBO

To verify that the targets relative to the figurative meaning could be primed bythe idiomatic expression, 15 subjects were tested in a priming experiment. Sub-jects were presented with the 16 idioms selected in the first phrase plus 5 newidiomatic expressions, plus 11 nonidiomatic phrases of the same length as theidioms. These phrases served as primes in a lexical decision task. The targetwords were formed by the 32 words relative to each idiom, plus 16 nonwords.Two lists were made. Each list included 16 target words, either related or unrelat-ed to each idiom, plus 16 pronounceable nonwords. Every subject saw all thepriming idioms only once, and the correspondent targets either were related tothe figurative meaning or were the appropriate controls. The results showed thatthe idioms effectively primed the related targets, with respect to the controls: Themean reaction times (RTs) were 556 and 606 ms for related and control targets,respectively [F'(l, 14)= 11.5,p< .01 by subjects, F"(1, 15) = 6.8,p < .02byitems].

Two sentences were constructed for each idiomatic expression, one biasingthe literal meaning, the other biasing the figurative meaning (see Appendix). Anattempt was made to avoid, as far as possible, including words in the sentencesthat were associated to the related target, so that the bias toward one type ofmeaning was produced at the interpretive level of analysis of the sentence, not bylexical preactivation.

Two tapes were formed. Each tape included, in a different order, 16 sentencescontaining the idiomatic expressions, either in the literal or the figurative contextbias version, randomly intermixed with 25 filler nonidiomatic sentences. Eachtape was preceded by eight sentences for practice. Four lists of visual targetswere paired with each tape. Each list included 16 experimental target words, 8filler words, 17 filler pronounceable nonwords, plus 4 words and 4 nonwords tobe used as practice. The 16 experimental target words included four wordsliterally related to four different idioms, four words figuratively related, and eightcontrol words, all relative to different idioms, so that every subject was presentedwith each idiom only once, paired with a target word in one of the differentexperimental conditions. The two groups of 20 subjects were assigned randomlyto the two experimental conditions determined by the type of bias produced bythe sentence context, literal or figurative. Within each group of 20 subjects, foursubgroups were determined by the association with one of the four lists oftargets.

Procedure

The sentences were recorded with a Revox B77 tape recorder by a male speakerwith a normal intonation. The tapes then were digitalized by means of a digit-to-analog converter, and the resulting output was analyzed to find the exact offset ofthe last word of the idiom. A pulse of about 5 volts was located in correspon-dence with the offset. The tape recorder was connected to an Apple lie micro-

8. COMPREHENSION OF AMBIGUOUS IDIOMS IN CONTEXT 169

computer that produced the display of the target word on the computer screen andrecorded RTs and errors.

The subjects were seated in front of the computer screen, with their right andleft fingers resting on the 9 and 5 buttons of the computer console, and listened tothe sentences through the headphones. Subjects were instructed to listen to thesentences carefully for comprehension, and that they would be given a list ofsentences for a recognition test at the end of the session. They also were told tolook at the screen where a string of letters, either word or nonword, sometimeswould appear. As soon as they detected it, they had to make a lexical decision,pushing the relative buttons, as fast and as accurately as possible. When theexperiment was finished, the subjects were presented first with the recognitiontest, and then with a list containing the idioms used in the experiment. They hadto mark those items they were unfamiliar with or did not know the meaning of.This procedure of checking the familiarity of the idiomatic meanings was fol-lowed in all of the experiments in the present chapter.


An analysis of the protocols of the recognition test showed that subjects hadunderstood the task correctly, listening for comprehension, as instructed. Thedata consisted of mean RTs of correct responses. Outliers that were +2 standarddeviations (SDs) from the mean for each subject (2.2% for the literal contextgroup, 1.5% for the idiomatic context group) were substituted by the mean persubject through all conditions. RTs exceeding 1,400 ms were cut off (.9% for theliteral, .6% for the idiomatic group). Also the data relative to the idioms thatwere not known to the subjects were discarded (3.1% for the literal, 1.8% for theidiomatic group). Finally, 1.5% of the data for the literal group and .9% for theidiomatic group were cut off due to failure of the equipment.

Mean response times for the two types of context are shown in Table 8.1. Ananalysis of variance (ANOVA), with one between-subject (context) and twowithin-subject factors, type of prime-target relation (literal, figurative), and typeof target (experimental, control), was carried out on the RT data. In the analysisby subjects, the significant factors were type of relation, F'(1, 38) = 4.05, p <.05, and type of target, F'(l, 38) = 15.10, p < .001. The interaction betweenthe two factors was also significant, F'(l, 38) = 5.69, p < .05.

In the analysis by items, context was not significant, F"(1, 30) = 2.84 p <10, and neither was type of relation, F"( 1, 30) = 1.07. The factor type of target

was significant, F"(1, 30) = 12.7, p < .001. The interaction between type oftarget and type of relation was not significant, F"(1, 30) = 1.95.

The data relative to the group that was presented with the literal context wereanalyzed separately. RTs to experimental targets, that is targets related to eitherthe literal or figurative sense of the idioms, were facilitated with respect to RTs tothe control condition [F'(l, 19) = 7.6 p < .05; F"(l, 15) = 9.6, p < .01]. The

170 COLOMBO

TABLE 8.1Mean Lexical Decision Times and Error Percentage

(in Parentheses) in Experiment 1

Context

Literal Figurative

Type of Target Experim. Control Experim. Control

Prime-target RelationLITERAL 639 (0) 716 (.6) 692 (0) 760 (.6)FIGURATIVE 708 (1.5) 732 (0) 725 (.6) 774 (.6)

interaction between type of prime-target relation and type of target was signifi-cantly by subjects, [F'(l, 19) 4.7, p < .05; F"(l, 15) = 2.4, p > .1], showingthat only the 77-ms facilitation for targets related to the literal meaning of theidioms (kick the bucket-pail) was significant as tested by the Newman-Keulsstatistics (p < .001), whereas the difference between targets related to the figur-ative reading of the idiom and its control did not reach significance. The othersignificant result from the ANOVA related to the factor type of prime-targetrelation [F'(l, 19) = 8.5, p < .01; F( 1, 15) = 2 , p > .1], which showed thatliterally related targets were faster than figuratively related targets. Error rate wasvery low, 2.2% of the total set of data.

The data relative to the idiomatic context group also was subjected to the sameanalyses. As for the former group, the factor type of target was significant [F'(l,19) = 7.5, p < .05; F'( 1, 15) = 5 7 p < .05], but the interaction this time didnot approach significance. The rate of errors was 1.8%.

The lack of a significant interaction between context, type of relation, andtype of target in Experiment I indicates that priming can be obtained for targetsrelated to both the literal and the figurative meaning of the idiom. The primingeffect for the figuratively related target words is smaller, independent of theinfluence of context. In fact, when analyzed in the literal context condition only,this priming effect failed to reach significance. This result could be due just tolack of power. It also could be explained assuming that the activation of theidiomatic meaning takes more time, with respect to the activation of the literalmeaning, and so has not reached asyntote when priming is measured at the offsetof the idiom. The bias offered by the idiomatic context could be sufficient tospeed up activation.

Another possibility cannot be discounted. That is, targets related to the literalmeaning of the noun were associated to it, whereas targets related to the figur-ative meaning were related semantically to the figurative interpretation of thewhole idiom. Semantic relations seem to produce more fragile priming effectsthan do associative relations (Colombo & Williams, 1990; Lupker, 1984).


The fact that in the literal context, RTs to targets related to the figurativemeaning were facilitated, although to a smaller degree, seems to be in oppositionto the findings of Cacciari and Tabossi (1988). In their experiments, when unpre-dictable idioms were used in neutral contexts, there was no evidence supportingthe view that the figurative meanings of idioms had been activated immediately.Only after 300 ms was there evidence of priming for the figurative meaning.Although the context was neutral, the idiomatic reading of the phrase was theonly one possible, because the idioms did not have a literal interpretation. There-fore the idiomatic interpretation had to become active at some point. Thus, itappears that the language processor first had considered the literal interpretationof the idiomatic phrase, and had rejected it only when it did not produce aplausible or meaningful interpretation. However, with more predictable idiomsthe figurative meaning was activated, whereas the literal word meaning was not.

A possible explanation for the inconsistency of the present results with thoseobtained by Cacciari and Tabossi (1988) is in terms of idiosyncratic differencesamong idioms. Perhaps the idioms used in Experiment I of the present study,although they have a literal interpretation, are quite common in their idiomaticreading, which was therefore available relatively quickly. Another differencebetween the two studies is in the relative proportion of filler sentences andidiomatic sentences (16 to 25 in the present experiment, 12 to 60 in Cacciari &Tabossi). The higher proportion of sentences containing idioms in the presentstudy may have created an expectation that activated the idiomatic meaning, eventhough it was not biased by the context.

The present results also can be explained in terms of an interpretation thatlocates the effect of facilitation on the integration process that occurs after wordrecognition, rather than on processes occurring before the presentation of thetarget. Facilitation of the target word could be produced by its rapid integrationin the ongoing mental representation being constructed. It has been suggestedthat the lexical decision task is particularly sensitive to postaccess integrationprocesses (Balota & Lorch, 1986; Colombo & Williams, 1990; Forster, 1981;Lupker, 1984; Seidenberg, Waters, Sanders, & Langer, 1984). Judgment ofwhether a string of letters is a meaningful word or not is claimed to be sensitiveto the presence of a semantic relation between prime and target, so that itbecomes easier to respond when the target word is easily integratable into theformer context. On this account, facilitation of target words depends on the easeof constructing a representation congruent with the preceding information. Whenthe figurative sense of the idiom is activated, even in the literally biased context,it might constitute a local context against which the target word is integrated, andcapable therefore of producing facilitation.

The target words related to the literal meaning appeared to be activatedindependently of context. As Gibbs (1986) rightly pointed out, however, this factdoes not imply necessarily that the literal interpretation of the sentence as awhole is being constructed. The process of constructing the literal interpretation

171

172 COLOMBO

of a sentence is a more complex process compared to that of retrieving a meaningstored in memory, as seems to be the case with idioms. Thus, one very wellmight argue that, although the literal interpretation always is initiated, it isabandoned as soon as the idiomatic meaning is retrieved and found to be con-gruent with the context. Moreover, there is another reason why facilitation ofliterally related targets cannot be taken as evidence in favor of the contextuallyindependent construction of literal meanings. Namely, facilitation could be attri-buted to backward priming (Koriat, 1981; Seidenberg et al., 1984; Van Petten &Kutas, 1987). As prime words and literally related target words were associated,and were presented at 0 Stimulus Onset Asynchrony (SOA), backward primingmight well have occurred. Therefore, the present results are inconclusive withrespect to the problem of whether the literal interpretation always is computed.

EXPERIMENT 2

In Experiments 2 and 3 the course of activation of the idiomatic meaning, and ofits integration with the context, was investigated in a self-paced reading task. Inthis task, sentence segments of different lengths are presented on a screen and thesubject has to press a button when she or he has completed reading the segments.RTs are assumed to reflect processes involved in both identification and semanticactivation of single words, and in higher level comprehension processes. The useof this paradigm should allow examination of the time course of activation ofboth the literal and the figurative meaning.

It was assumed that, while the segments of the sentence accumulate, a struc-tural interpretation is constructed. If, at presentation of the idiom, the figurativemeaning has been activated, two incompatible interpretations would be active,and the language processor has to select one of them. Decisions on whichinterpretation to select, and when to operate the selection, may depend on severalfactors. One of these factors is contexlI. If the context is neutral, the languageprocessor might decide to delay selection until disambiguating information isprovided. Or, it might select immediately which interpretation is more likely onthe basis of frequency of occurrence (e.g., how often an idiom is used in thefigurative sense). When the context is biasing one interpretation, then an immedi-ate selection can be made (Frazier & Rayner, 1990; Rayner & Frazier, 1989).

If in the condition where the context is biasing the figurative.meaning infor-mation that is not congruent with the context is provided after the idiom ispresented, a potential garden path is created. Thus, if a delay is found in encoun-tering this incongruous information, th is should be evidence that selection of theidiomatic interpretation already had been made.

Let us consider the following example: "The man was going to kick thebucket, that was on the kitchen floor."


In this example the idiom is followed by a relative pronoun as subject of averb phrase (e.g., ". . . kick the bucket, that was . . ."). There are two possiblereferents to the pronoun. One is the noun of the idiom, the second is the wholephrase, considering the verb as a nominal. In the latter case, in Italian, therelative pronoun is preceded most often by the article the when it refers to awhole phrase. Therefore the favored referent in the example should be the nounof the idiom. This could be intended in the literal or in the figurative meaning.Some idiomatic expressions, although not each to the same extent, would allow apronoun to refer to the figurative meaning of its noun. Indeed several authors(Cacciari & Glucksberg, 1991; Gazdar, Klein, Pullum, & Sag, 1983; Gibbs &Nayak, 1989; Wasow, Sag, & Nunberg, 1983) argued that some idioms aresubject to internal modification by adjectives or relative clauses, which can applyto the whole idiom or to a part, and claimed that the meaning of an idiom can be acompositional function of its parts, which are not intended literally however.This view has led to the formulation of the idiom decomposition hypothesis(Gibbs & Nayak, 1989; Gibbs, Nayak, & Cutting, 1989).

In Experiments 2 and 3, the context of the sentences biased either the literal oridiomatic meaning or was neutral. The idiom always was followed by a relativeclause in all of the contexts. The relative clause was formed by the relativepronoun (che, "that"), plus a verb followed by either a noun phrase, a preposi-tional phrase, or an adverb, which completed the verb phrase introduced by therelative pronoun, as in the following example: ". . . break the ice, that was inthe fridge." The relative pronoun referred to the literal meaning of the last wordof the idiom, which was always a noun. The prepositional phrase "in the fridge"contributed to the specification of the literal sense of the noun of the idiom, andconsequently also of the relative pronoun. If, at presentation of the relativeclause, the idiomatic meaning has been activated and integrated, reading thesuccessive segments containing the literally related clause should produce asituation similar to that created by a garden path, where the language processorhas to reconsider alternative interpretations to that formerly biased by the con-text. The garden path thus should produce a delay in reading times. If, on theother hand, the figurative meaning had not been activated, or already had beensuppressed, no delay would be expected on the postidiom region.

Sentences were presented visually, divided into segments of different length,formed by one or more words. The first critical segment was formed by theidiomatic phrase, the second by a relative pronoun as subject of the relativeclause, and the third by a verb phrase. The segments corresponding to the idiomand the region postidiom were formed by the same words, as shown in ExamplesI and 3. In Experiment 2 the context was either ambiguous, or biasing the literalmeaning of the idiom, as shown, respectively, in Examples I and 3:

1. It looked/like/the man/was going/to kick the bucket/that/was on thekitchen floor/.

173

174 COLOMBO

2. It looked/like/the man/was going/to hit the bucket/that/was on the kitch-en floor/.

3. The man/was enraged/and/in an outburst/he kicked the bucket/that/wason the kitchen floor/.

4. The man/was enraged/and/in an outburst/he hit the bucket/that/was onthe kitchen floor/.

In the control condition, the verb of the idiom was substituted by another verbin order to form a phrase with a meaning similar to that of the literal meaning ofthe idiom, but which could not be intended idiomatically (Sentences 2 and 4).Reading times for the idiomatic string versus the control string were compared inthe two contexts.

The predictions depend on whether it is assumed that, besides activation ofthe meanings, integration and selection also occur during the time interval be-tween presentation of the critical segment and the pressing of a button. Thepredictions also may depend on the relative time course of activation of the twopossible interpretations.

If the figurative meaning, being associated to the whole phrase, is activatedautomatically, one might not expect any lengthening in reading times due purelyto activation, but only to the further operations of integration and selection ofmeanings. It can be assumed that these operations are carried out immediatelyonly when incompatible interpretations are active. This assumption is consistentwith one made by Frazier and Rayner (1990) in a study where the processesinvolved in the resolution of meaning, as opposed to sense ambiguity, wasinvestigated. In the first type of ambiguity the meanings are unrelated (i.e.,"river bank" vs. "money bank"), whereas in the second the two meanings are tosome extent related (i.e., "newspaper," "magazine" vs. "publisher"). Accord-ing to the data found by Frazier and Rayner, the language processor seems toemploy different strategies in the two cases. In the resolution of meaning ambi-guity the processor has to decide between incompatible interpretations, but whenthe meanings are related, as is the case for sense ambiguity, the two interpreta-tions may be to some extent compatible.

The selected idioms were understandable and plausible in both the literal andthe figurative interpretation. Considering first the idiom region, when the idiomis in a neutral context, if both the literal meaning and the figurative meaning areactivated almost simultaneously one would expect a delay in reading times withrespect to the control condition, due to the necessity of a selection betweenincompatible interpretations.

When the context biases the literal interpretation, even if the idiomatic mean-ing has been activated, the language processor may select immediately betweenthe alternative interpretations. Thus, a lengthening of reading times with respectto the control condition would be expected in the idiom region, due to the


selection process. However, if the idiomatic meaning has not been activated, orhas been activated too late, when the literal interpretation already has beenconfirmed, no difference would be expected between the experimental and thecontrol conditions.

It is assumed that if the figurative interpretation is selected, when the relativeclause is encountered (second and third segments) a garden path occurs. Thus,reading times in this region should give information about whether the figurativemeaning has been activated, and when. Longer RTs in the postidiom region withrespect to the control condition were expected in the sentences with a neutralcontext, showing that the figurative meaning has been activated. If a delay isfound in the literal bias context, this would indicate that the figurative meaningbecomes active independent of context.

Method

Subjects. Twenty-eight students of the University of Padua volunteered assubjects in the experiment.

Material. Sixteen idiomatic phrases were selected for the experiment. Theywere all ambiguous idioms formed by a verb phrase and a noun phrase, with aplausible and meaningful literal interpretation, beside the figurative one. Nine ofthe 16 idioms had been used in Experiment 1; the remaining idioms werechanged because of the experimental condition's constraints, by which they hadto be modified by a relative clause.

Each idiom was preceded either by a neutral context or by a context biasingthe literal interpretation, forming the experimental condition of context bias.Moreover, a control condition for each type of context was created, in which theverb of the idiom was changed with a synonym, or quasisynonym, so that theidiomatic meaning was not retained. The frequency of the verbs in the twoconditions was matched (72 for the verb of the idiom, 68 for the control verb).Thus, two factors were manipulated, context (neutral vs. literal bias) and type ofstring (idiom vs. control). The sentences are listed in the Appendix.

For each idiomatic phrase there were four experimental conditions, for a totalof 64 test sentences. The sentences were assigned to four lists. In each list eachidiom appeared once in only one of the four experimental conditions. A set of 49filler sentences was added to each list. None of the fillers contained idiomaticphrases. The order of the idioms was fixed within a list, but was differentbetween the lists. Each subject was presented with a list containing the 16experimental and 49 filler sentences, for a total of 65 sentences.

The sentences were presented in segments of different length in order not tocreate an expectation for a particular type of chunk (word or phrase). The criticalsegments, on which the analyses had to be carried out, were the following: theidiom phrase, the relative pronoun, the region following the pronoun. These

175

176 COLOMBO

segments were exactly the same in all four conditions, except for the changing ofthe verb of the idiom in the control conditions.

Procedure

Sentences were displayed with a self-paced technique. Segments (one- to four-words long) were displayed in the center of the screen of a monitor commandedby an IBM computer, and remained there until the subject pressed the plus buttonon the numeric keypad. The software for executing the experiment and analyzingthe data was the MEL (Micro Experimental Laboratory). In the instructions givento the subjects particular emphasis was placed on sentence comprehension. Sub-jects were told that they would be asked questions about the sentences, to ensurethat they were reading for understanding. After each segment was displayed,subjects read it and pressed a button as soon as they thought they had understood.After the final segment appeared a blank was displayed. At random intervals,fixed within a list, a pause was made and the experimenter asked a questionabout the immediately preceding sentence. The data relative to these questionswere not analyzed, as subjects were almost totally correct. RTs were recorded,from the appearance of each segment on the screen, to the pressing of the plusbutton.


The data consisted of mean RTs for critical segments. Data relative to idioms forwhich the idiomatic interpretation was unknown to a subject were discarded(6.7%). Also, 3.2% of the data were missing because of a failure of the equip-ment. Separate ANOVAs were carried out on the mean RTs of the three criticalsegments, with two factors, two levels each (neutral vs. literal context; idiom vs.control).

None of the factors, or interactions, was significant. For the first segment themean reading times were respectively 862, 898, 866, and 880 ms, for the neutralsentences containing the idiom, for their controls, for the literal sentences con-taining the idiom, and their controls (all Fs < 1). For the second segment (therelative pronoun) the factor context was not significant, F < 1, the factor idiomversus control was marginally significant, F(l, 27) = 4.15, p = .051, and sowas the interaction with context, F(l, 27) = 3.5, p > .05. Mean RTs for therelative pronoun were 638, 716, 671, and 661, respectively. An analysis of thesimple main effects showed that RTs for the relative pronoun were shorter inthe sentences containing an idiom in the neutral context, with respect to thecontrol, F(l, 27) = 5.56 p < .05.

For the third and final segment (containing the remaining part of the relativeclause) reading times were 949, 978, 897, and 911 ms, and no factor wassignificant.


According to these data, there is no evidence for activation of the figurativemeaning in either the literally biased context, because no delay was found inreading the idiomatic string with respect to the control, or in the neutral context.The only evidence of an effect of the presence of an idiom was in the faster RTsfor the relative pronoun region when the idiom followed a neutral context, inexactly the opposite direction to what would be expected if the language pro-cessor were facing a garden path, a result that does not have an obvious in-terpretation. If this result reflected the fact that the idiomatic interpretation hadbeen retrieved and integrated, then longer RTs should have been found at least inthe final region, where a literal interpretation clearly is required.

Considering all three segments, these results suggest that the language pro-cessor has a preference for the literal reading of idiomatic expressions when nofigurative bias is given. It is worth pointing out that in this experiment theidiomatic phrases never were biased by an idiomatic context, because if such abias had been given, the sentences would have become implausible when therelative clause, referring to the literal meaning of the noun of the idiom, waspresented. Given the lack of any significant effect in the expected direction, andtherefore the weakness of a conclusion based on a null result, the claim that thefigurative meaning was not activated should be qualified by a result showing thatindeed, when the idiomatic meaning is activated, a garden path occurs, with aconsequent delay in RTs.

EXPERIMENT 3

In Experiment 3, two types of sentence contexts were manipulated, a literalcontext and an idiomatic context. The literally biased sentence formed a plausiblesentence, whereas the sentence with a figurative bias became implausible whenthe relative clause was encountered, as shown in Examples 5 and 6:

5. Giovanni was so enraged that he wanted to hit something and he kickedthe bucket, that was on the kitchen floor.

6. The rich old man was very ill, and his inheritor was hoping that he kickedthe bucket, that was on the kitchen floor.

When the context is disambiguating, if the processor computes both the literaland the figurative interpretation, the ambiguity could be resolved immediately. Inthe condition where the context is biasing the literal interpretation of the idiomthe sentence is plausible, and there is no garden path. This condition thereforewas used as a control for both the idiom region and the region following theidiom. If the context biases the idiomatic meaning, a garden path must occurwhen the relative clause is encountered. Thus, it is expected that reading timesfor the postidiom region should be faster in Example 5 than in 6.

177

178 COLOMBO

In order to distinguish the longer reading times due to the effect of im-plausibility, from those due to a garden path, a third condition was created. Thenoun of the idiom was replaced by another noun that did not form an idiomaticexpression, but made the sentence implausible (Example 7).

7. The rich old man was very ill, and his inheritor was hoping that he kickedthe pail, that was on the kitchen floor.

In the sentence fragment ". . . kicked the pail, that was on the kitchen floor,"the implausibility starts when the word pail is presented, given the precedingcontext. At this point, if there is an effect of implausibility, there should be adelay with respect to the corresponding segment in Examples 5 and 6, where thenoun of the idiom is perfectly acceptable as a continuation in both contexts.However, the effect of implausibility should be smaller with respect to a garden-path condition, because the language processor can decide immediately that thesentence is implausible, and does not have to consider alternative interpretations.

Method

Subjects. Thirty students of the University of Padua volunteered as subjects.

Material. To the set of idioms used in Experiment 2, two more idioms wereadded, in order to balance the number of items within each condition. Thecritical regions were formed by the idiom and the postidiom regions, and weremodified with respect to Experiment 2. The idiomatic phrase was divided intotwo segments. The first segment was formed by the verb of the idiomatic verbphrase (kick), the second segment by the noun phrase following the verb (thebucket). This modification with respect to the presentation conditions of Experi-ment 2, where the idiom had been presented as a single segment, was made inorder to control for differences in activation of the different parts of the idiom. Italso made it possible to verify whether a delay would be found in the controlcondition, as the implausibility was made apparent by the noun of the verbphrase. The third segment included a relative pronoun as subject, and its verb,whereas the fourth segment was formed by a prepositional phrase or an adverband ended the sentence. The four segments formed the critical regions on whichanalyses of reading times were performed.

Eighteen nouns were selected to substitute for the nouns of the idioms in thecontrol condition. They were comparable to the nouns of the idioms for frequen-cy and length, (mean frequency = 41.8 for the idiom noun, 32.6 for the control;mean length = 6 for the idiom noun, 5.3 for the control.) The 18 idioms wereinserted in sentences with a literal ot an idiomatic bias, for a total of 54 testsentences. Three lists were created, each containing an equal number of testsentences (18) plus 53 filler trials. In each list each idiom appeared in only one ofthe three experimental conditions.


Because sentences in this experiment could be plausible or implausible, someof the filler sentences used in Experiment 2 also were made implausible, in orderto have an equal number of plausible and implausible sentences within each list.Thus, each subject was presented with 71 sentences.

Procedure

The procedure was similar to that used in Experiment 2, except for the followingchanges. The self-paced reading task was modified by the introduction of asecondary task. Subjects were asked to perform the self paced reading task,pressing a button with the index finger of their right hand, and to decide asquickly as possible whether the sentence made sense, pressing a button for a"yes" response and another for a "no" response, with the index and middlefingers of their left hand. The buttons used in the sentence decision task weredifferent from the button used for self-paced reading.

Once again, reading for comprehension was stressed particularly, however nocomprehension tests were given. Subjects were required to make the plausibilitydecision as fast as possible, and in order to do that, they were told that they had toread and understand the segments as soon as they were presented.

Results

For the reading times, separate analyses were carried out for each of the foursegments in the three experimental conditions. RTs for idioms that were un-known to subjects in their figurative interpretation were discarded (6.6% of thedata). Also RTs for sentences incorrectly classified were removed (16%) andwere analyzed as errors. Mean RTs are displayed in Table 8.2.

The ANOVA on the first and second segments, corresponding to the verb andto the noun of the idiom, respectively, showed that there was no significantdifference in reading times, F(2, 58) = 1.29, F(2, 58) = 1.69, respectively (seeTable 8.2).

The analysis on the third segment showed a marginally significant effect, F(2,58) = 2.57, p < 1. T tests on the means showed that the difference between theidiomatic test sentence and the literal sentence was significant, (p < .05). Final-ly, the analysis on the last segment was significant, F(2, 58) = 4.36, p < .05. Ttests showed that the RTs to the final region in the idiomatic context conditiondiffered from RTs in both the idiomatic control condition and literal contextcondition, p < .05. The latters did not differ from each other. Item analyses werenot significant for any segment.

An analysis also was carried out on the mean RTs to decide whether thesentence was plausible or implausible. Mean RTs were 609 ms for the idiomaticcontext condition, 586 ms for the control condition, and 680 ms for the literalcontext condition. The difference were not significant, F(2, 58) = 2.5, p < 1.Also the analysis by items showed a marginal effect, F(2, 34) = 2.48, p < .1.

179

180 COLOMBO

TABLE 8.2The Mean Reading Times for the Idiom (First Two Rows) and thePostidiom Regions (Third and Fourth Rows), the Mean ReactionTimes for Correct Plausibility Decisions, and Error Percentage

(in Parentheses, Fifth Rovw) for the Different ExperimentalConditions in Experiment 3

Idiom. ConteK1 Idiom. ContextExperimental Control Literal Context

1 idiom 679 673 6572 idiom 756 697 7101 postidiom 719 689 6782 postidiom 762 706 712Plausibility decision 609 (3.3) 586 (3.3) 680 (9.6)

In the analysis of errors, it was found that there was a significant trend forerrors to occur more frequently in the literal context condition, F(2, 58) = 14, p< .001. Error percentages were 3.3% for the idiomatic context, 3.3% for thecontrol condition, and 9.6% for the literal context condition. T tests on the meansshowed that the literal context condition was significantly different from theother two conditions, which did not differ from each other (p < .01).

To interpret the present results it is necessary to assume that the most likelyreferent for the relative pronoun is the noun of the idiom, rather than the idiom asa nominal. Because this assumption is crucial, it was advisable to have someempirical grounds to support it. Thus, a test was carried out on the material ofExperiment 3 to verify the probability that the relative pronoun was interpreted asreferring to the literal meaning of the noun of the idiom, to its figurative mean-ing, or to the whole phrase as a nominal. The sentences of Experiment 3 weredivided into two lists. One list contained 18 sentences, half of which wereformed by the idioms in the figurative context and half by the idioms in the literalcontext. In the second list the order was reversed, so that subjects saw all theitems only once. Two more lists were constructed, containing the same idiomspreceded by a neutral context, plus filler sentences in which the noun of theidiom had been substituted by another noun, as in the control condition ofExperiment 3. The sentences ended immediately after the relative pronoun, andsubjects were asked to provide a meaningful completion. Four groups of 10subjects each were used, randomly assigned to one of the lists.

The data for four sentences were lost due to an error of the experimenter. Thecompletions were scored according to the following categories. The first catego-ry included completions where the relative pronoun referred to the literal mean-ing of the noun. For the second and third categories the relative pronoun referredto the figurative meaning of the noun, or to the whole phrase, respectively. Thefourth category consisted of completions that were paraphrases of the meaning ofthe idiom, or other types of nonrelevan: completions.


It was found that, for the sentences biasing the idiomatic meaning of theidiom, 51% of completions were formed by clauses in which the relative pronounreferred to the figurative meaning of the noun of the idiom, 17% by clausesreferring to the literal meaning, and 9% by clauses referring to the idiom as awhole. The fourth category (nonrelevant completions) included 23% of the com-pletions. For control sentences, where the noun of the idiom has been replacedby another noun to form a control condition, completions in terms of the literalmeaning of the noun were 95.5%; the remaining completions belonged to thecategory of paraphrases or other nonrelevant completions. For sentences with aneutral context, 60% of the completions referred to the literal meaning of thenoun, 2.5% to the whole idiom, 7% to the figurative meaning of the noun, andthe remaining were classified as paraphrases or other.

Clearly, these results show that subjects tended to assign the noun of the idiomrather than the whole phrase as a referent to the relative pronoun. Moreover,when the bias of the context constrains a figurative interpretation, subjects triedto make a meaningful completion by proposing an interpretation of the relativeclause that is congruent with the context, and refers to the figurative meaning. Incontrast, when the context is neutral the preferred referent was the literal mean-ing of the noun.

Reference to the noun occurred more easily when idioms are transparent. Forinstance, for the idiom break the ice the connection between literal and figurativemeaning is quite transparent, and 80% of the completions in the present test wereformed by phrases referring to the figurative meaning of the noun, 10% to theliteral meaning of the noun, none to the idiom as a whole. With nontransparentidioms, like mangiare la foglia ("to eat the leaf"), meaning something like "tograsp the deceit," no reference to the literal or figurative meaning of the nounwas given.

Discussion

The results of Experiment 3 show that, when the context biases the idiomaticmeaning, RTs in both the ambiguous region and in the postambiguity regionincrease. Such increase is only marginally significant for the ambiguous region.This might be due to variability in the extent to which the figurative meaning isactivated and integrated with the context. In contrast, no difference is found forthe two conditions in the first part of the idiom (first segment). Apparently, theidiomatic meaning is not activated at such an early test position, or, alternatively,there is no cost to the activation.

The occurrence of a garden path in the postidiom region shows that theidiomatic meaning has been activated and integrated in the context, but only inthe biasing context, when a reanalysis of the interpretation has been required.When the context is neutral, as was the case in Experiment 2, there is nolengthening of reading times in the ambiguous region, nor is there any evidenceof the occurrence of a garden path in the postidiom region. Moreover, no length-

181

182 COLOMBO

ening of RTs for the ambiguous region appears when the context is biased towardthe literal interpretation.

It is worth noting that no significant delay was found in the control conditionof Experiment 3, relative to the literal condition. That is, processing the controlnoun, which made the sentence implausible, in the verb phrase did not takesignificantly longer with respect to the literal condition, in either of the threesegments. It appears that when the implausibility is very obvious, the languageprocessor immediately decides that the sentence is not plausible and this decisionis not costly, in processing terms.

What seems to involve more complex decision processes, reflected in longerRTs, is the condition in which there is an ambiguity and two interpretations arepossible, one on the basis of the context preceding the idiom, and the second onthe basis of information following it. The language processor seems to prefer tomaintain the interpretation that had been selected initially on the basis of thecontext (the figurative interpretation in the case of the idiomatic context). Thisfact is confirmed by the results of the completion test where 51% of the comple-tions assigned the figurative meaning of part of the idiom as a referent for therelative pronoun, even at the cost of grammatical correctness in some cases.However, such preference only occurs when the context biases the idiomaticmeaning. When the context is neutral the literal meaning of the noun is thepreferred referent, as shown by the completions in which the pronoun wasassigned a literally intended referent in 60% of the cases. Therefore, the con-struction of an interpretation of the relative pronoun in terms of the figurativemeaning of the noun conflicts with the latter preference. This conflict is reflectedin the RTs to the region immediately following the idiom.

Finally, when the whole relative clause has been read, its only possible in-terpretation, in terms of the literal meaning of the noun, compels the languageprocessor to reconsider the given interpretation, which leads to a decision ofimplausibility. It is interesting to note that the RTs to the sentence classificationtask do not reflect any differences. Presumably, the decision about the plau-sibility of the sentence is made before the end of the sentence. However, evenmore interesting, a higher number of errors has been found in the literal contextcondition with respect to the other two, despite the fact that it required a positiveresponse. It appears that the presence of an idiom increases the probability ofmaking an error in a literally biased sentence. This fact suggests that the idi-omatic meaning is active at the time of the sentence classification, and interfereswith the judgment. Perhaps, the fact that the effect of this activation is onlymanifest in the pattern of errors during sentence classification reflects a delayedactivation, due to postperceptual processes. Alternatively, it could be that theidiomatic meaning has been activated during the presentation of the sentence, butonly on some trials, and to an extent not detectable in selfpaced reading.

One more point to consider relates to the fact that in an idiomatic context theidiom can be perceived as modified by a relative clause referring to the figurative


meaning of one of its part (the noun). This is in agreement with the suggestionthat the figurative interpretation can be constructed, to some extent, on the basisof the different components of the idiom (Gazdar et al., 1983; Gibbs & Nayak,1989; Gibbs et al., 1989; Wasow et al., 1983). This view is in opposition to theidea that the idiomatic meaning is a stored representation directly associated withthe whole phrase, and might hold particularly for certain types of idioms, likebreak the ice, so-called "analyzable-transparent" in the classification proposedby Cacciari and Glucksberg (1991). For these types of idioms the construction ofan idiomatic interpretation may involve similar processes to those required toform an interpretation of unambiguous nouns in selecting the sense appropriate tothe context. For instance, the action of "breaking," rather than being applied to aphysically concrete domain, is mapped onto an abstract domain that defines ametaphorical connotation of the concept of "coldness" implied by the idiom.

The results of Experiments 2 and 3 suggest that with ambiguous idioms thepresence of a biasing context is a necessary condition for activation of thefigurative meaning to occur. This result is in opposition with that found inExperiment 1 where a different experimental paradigm had been used. In thatexperiment, priming was found for both the figurative and the literal meaningindependent of context. A similar conflict was found elsewhere in the literature,with respect to ambiguous words. Data provided by the cross-modal primingparadigm (e.g., Onifer & Swinney, 1981; Swinney, 1979) showed a differentpattern from those obtained using the eye fixations methodology (Rayner &Frazier, 1989). As has been noted in the discussion of Experiment 1, performinga lexical decision task in the cross-modal priming paradigm can involve mecha-nisms that are peculiar to the specific experimental situation, and so may bedifferent from those involved in another situation, requiring a different task.Thus, it could be that priming effects are magnified in the priming paradigm, andby the decision mechanism involved in lexical decision. In self-paced reading(and may be also in normal reading) concepts may be activated just very briefly,and to a smaller extent, not sufficient to be detected. If this is the case, thepriming paradigm, especially when the lexical decision is used, is more appropri-ate to investigate how concepts are represented and related, rather than theprocesses involved in reading and listening.

CONCLUSION

Several factors are involved in the comprehension of idioms. One of thesefactors, already pointed out by Cacciari and Tabossi (1988), is predictability.This factor itself presumably involves several components, which now are con-sidered.

An important component may be frequency of occurrence of an idiomaticexpression, that is, how frequently the idiom has been encountered, or how

183

184 COLOMBO

strong is the connection between the string of words forming an idiom and itsassociated meaning. The contribution of this factor is likely to vary, in view ofthe fact that idioms can be subject to syntactic operations. Thus, as the super-ficial form of the string is not always the same, the availability of the storedmeaning is likely to vary. Thus the hypothesis of an automatic activation shouldbe modulated, considering the relative strength of the connection between anorthographic or phonological form, word or phase, and its semantic representa-tion as predictor of the automaticity of the process involved. If there is a strongconnection, the meaning is more likely to become available, and in a shorter time(Cohen, Dunbar, & McClelland, 1990)

Another important component involved in the comprehension of idioms is thedegree of frozenness (Gibbs & Gonzales, 1985). Related to this aspect is thesyntactic form in which the idiom is expressed. Syntactic cues, like presence orabsence of articles, are certainly critical to determine the probability that anexpression is interpreted as an idiom, rather than literally. Moreover some idio-syncratic aspects of idioms also could be important. For instance, some idiomaticexpressions make reference to events that would be unlikely if intended literally,as, for instance, in the case of idioms like lose face, or, in Italian, idioms likeavere il pollice verde ("to have a green finger").

Context is obviously one of the most relevant variables to consider whenstudying the comprehension of idiomatic expressions. In particular, it seemscritical, especially when ambiguous idioms are involved, to understand whetherand how context interacts with other factors to determine the point at which theidiomatic meaning is activated and integrated. A distinction that may be useful inthis respect is that of local and global context. To illustrate this distinction a fewwords are necessary. The meanings of single words in a phrase or sentenceinteract both locally and globally. They must be combined locally in order todefine a semantic domain (see Examples 8, 9, & 10) where the context contrib-utes to the selection of the particular sense of the word run, that is, a concretesense involving a physical and spatial movement as in 8, a concrete spatial sensewith no real movement involved as in 9, or an abstract sense as in 10. At theglobal level the meanings of the single words must be integrated to form aninterpretation at the whole sentence level:

8. The train ran through the country.9. The scar ran through his face.

10. The business ran efficiently.

Thus, local context determines cohesion and integration of lower level constit-uents, like the phrase in which a word is included. Global context determines andis linked to higher level elaborative processes, like interpretation of the wholesentence.


To illustrate the relevance of this distinction between local and global levelsconsider the following examples. An ambiguous idiom, like break the ice, canhave a meaningful interpretation at the local level. However, an interpretation ofthe opaque unambiguous idiom shoot the breeze based on the compositionalmeanings of its words would not be meaningful, unless used in a peculiar sense.Alternatively, in a context biasing the idiomatic meaning of kick the bucket("die") (e.g., "The old man had been sick for a long time and finally he kickedthe bucket") the literal interpretation, possible at the local level, would not bemeaningful, or at least not very plausible, at the global level. Lack of a suc-cessful integration of the interpretation constructed at the local level with theglobal context therefore could act as a triggering device for the activation of thefigurative meaning.

Finally, one more factor to consider is transparency. As pointed out earlier, atransparent idiom may require different processing strategies from those requiredby an opaque idiom. Similarly, Frazier and Rayner (1990) showed that thelanguage processor uses different strategies to deal with words having differentmeanings or different senses. If this is the case, constructing an idiomatic in-terpretation might not, or not always, involve the simple activation of a storedmeaning, at least for transparent idioms, because it would be a compositionalfunction of the figurative meaning of its words, and the language processorwould have to use different strategies with different types of idioms. Such a viewwas suggested by Gibbs et al. (1989) and is consistent with their results.

In conclusion, the study of the processes involved in the comprehension ofidiomatic expressions appear to be central to the understanding of the humanlanguage processor.

ACKNOWLEDGMENTS

The preparation of this chapter was supported by a CNR grant in 1988 to theauthor. The data for the first experiment were reported at the Joint Conference ofthe Experimental Psychological Society, the Societa Italiana di Psicologia and theSocietA Italiana di Neuropsicologia held in Padua, April 1986. The data for thesecond and third experiments were reported at the Second Workshop on Lan-guage Comprehension held in Aix-en-Provence, April 1991. I wish to thankSandro Bettella for his invaluable aid with technical problems and programming,Emanuele Tressoldi for helping run the subjects in Experiment 1, and the Depart-ment of Electrotechnical Engineering of the University of Padua for technicalsupport in the preparation of the stimulus material for Experiment 1. I am alsograteful to Robert Pierce for editing the paper.

Address correspondence to: Lucia Colombo, Dipartimento di PsicologiaGenerate, P.zza Capitaniato, 3, 35139 Padova, Italy. Email address: PSICO04@IPDUNIVX. UNIPD . IT

185

REFERENCES

Balota, D. A., & Lorch, R. F. 1. (1986). D)eplh of automatic spreading activation: Mediatingpriming effects in pronunciation but not in lexical decision. Journal of Experimental Psychology:Learning, Memory, and Cognition, 12(3', 336-345.

Bobrow, S., & Bell, S. M. (1973). On catching on to idiomatic expressions. Memory & Cognition,1, 343-346.

Bortolini, U., Tagliavini, C., & Zampolli, A. i1 ?72). Lessico di frequenza delta lingua italianacontemporanea [Lexicon of the frequency of the contemporary Italian language ]. Milano, Italy:Garzanti.

Cacciari, C., & Glucksberg, S. (1991). Understanding idiomatic expressions: The contribution ofword meanings. In G. B. Simpson (Ed.). Understanding word and sentence (pp. 217-240). Amsterdam, Netherlands: Elsevier.


Chomsky, N. (1980). Rules and representations. New York: Columbia University Press.Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control of automatic processes: A

parallel distributed processing account of the Stroop effect. Psychological Review, 97(3), 332-361.

Colombo, L., & Williams, J. (1990). Effects of word- and sentence-level contexts upon wordrecognition. Memory & Cognition, 18(2), 153--163.

Duffy, S. A., Morris, R. K., & Rayner, K. (1988). Lexical ambiguity and fixation times in reading.Journal of Memory and Language, 27, 429-446.

Forster, K. 1. (1981). Priming and the effects of sentence and lexical contexts on naming time:Evidence for autonomous lexical processing. Quarterly Journal of Experimental Psychology,33A, 465-495.

Fraser, E. (1970). Idioms within a transformational grammar. Foundations of Language, 6, 22-42.Frazier, L., & Rayner, K. (1990). Taking on semantic commitments: Processing multiple meanings

vs. multiple senses. Journal of Memory and Language, 29, 181-200.Gazdar,G., Klein, E., Pullum, G. K., & Sag, I. A. (1983). Generalized phrase structure grammar.

Hertfordshire, England: Basil Blackwell.Gibbs, R. (1980). Spilling the beans on unders anding and memory for idioms in conversation.

Memory & Cognition, 8, 449-456.Gibbs, R. (1986). Skating on thin ice: Literal meaning and understanding idioms in conversation.

Discourse Processes, 9, 17-30.Gibbs, R. W., & Gonzales, G. P. (1985). Syntactic frozenness in processing and remembering

idioms. Cognition, 20, 243-259.Gibbs, R. W., & Nayak, N. (1989). Psycholinguistic studies on the syntactic behavior of idioms.

Cognitive Psychology, 21, 100-138.Gibbs, R. W., Nayak, N., & Cutting, C. (1989). How to kick the bucket and not decompose:

Analyzability and idiom processing. Journal of Memory and Language, 28, 576-593.Glass, A. L. (1983). The comprehension of idioms. Journal of Psycholinguistic Research, 12(4),

429-442.Koriat, A. (1981). Semantic facilitation in lexical decisions as a function of prime target associa-

tion. Memory & Cognition, 8, 587-598.Lupker, S. J. (1984). Semantic priming withotit association. Journal of Verbal Learning and Verbal

Behavior, 23, 709-733.Onifer, W., & Swinney, D. A. (1981). Accessing, lexical ambiguities during sentence comprehen-

sion: Effects of frequency of meanings and contextual bias. Memory & Cognition, 9, 225-236.Ortony, A., Schallert, D., Reynolds, R., & Antos, S. (1978). Interpreting metaphors and idioms:

Some effects of context on comprehension. Journal of Verbal Learning and Verbal Behavior, 17,465-477.

186


Rayner, K., & Duffy, S. A. (1986). Lexical complexity and fixation times in reading: Effects ofword frequency, verb complexity, and lexical ambiguity. Memory & Cognition, 14(3), 191-201.

Rayner, K., & Frazier, L. (1989). Selection mechanisms in reading lexically ambiguous words.Journal of Experimental Psychology: Learning, Memory, and Cognition. 15,(5), 779-790.

Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkovski, M. (1982). Automatic accessof the meanings of ambiguous words in context: Some limitations of knowledge-based process-ing. Cognitive Psychology, 14, 489-537.

Seidenberg, M. S., Waters, G. S., Sanders, M., & Langer, P. (1984). Pre- and postlexical loci ofcontextual effects on word recognition. Memory & Cognition, 12, 315-328.

Simpson, G. B. (1981). Meaning dominance and semantic context in the processing of lexicalambiguity. Journal of Verbal Learning and Verbal Behavior, 20, 120-136.

Swinney, D. (1979). Lexical access during sentence comprehension: (Re)consideration of contexteffects. Journal of Verbal Learning and Verbal Behavior, 18, 645-660.

Swinney, D. (1981). Lexical processing during sentence comprehension: Effects of higher orderconstraints and implications for representation. In T. Myers, J. Laver, & J. Anderson (Eds.), Thecognitive representation of speech (pp. 601-609). Amsterdam, Netherlands: North Holland.

Swinney, D., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journal ofVerbal Learning and Verbal Behavior, 18, 523-534.


Tabossi, P., & Cacciari, C. (1988). Context effects in the comprehension of idioms. In CognitiveScience Society (Ed.), Proceedings of the Tenth Annual Conference of the Cognitive ScienceSociety (pp. 90-96). Hillsdale, NJ: Lawrence Erlbaum Associates.

Tabossi, P., Colombo, L., & Job, R. (1987). Accessing lexical ambiguity: Effects of context anddominance. Psychological Research, 49, 161 -167.

Tanenhaus, M. K., Leiman, J. M., & Seidenberg, M. S. (1979). Evidence for multiple stages in theprocessing of ambiguous words in syntactic contexts. Journal of' Verbal Learning and VerbalBehavior, 18, 427-440.

Van Petten, C., & Kutas, M. (1987). Ambiguous words in context: An event-related potentialanalysis of the time course of meaning activation. Journal of Memory and Language, 26, 188-208.

Wasow, T., Sag, I. A., & Nunberg, G. (1983). Idioms: An interim report. In S. Hattori & K. Inone(Eds.), Proceedings of the XIII International Conference o]'Linguists (pp. 102-115). Tokyo:CIPL.

APPENDIX

Sentences biasing the literal (a) or the figurative (b) interpretation of the idioms(idioms are underlined in the figurative context) and target words, in parentheses(experimental control), used in Experiment 1. In italics are translations of theidioms or the English equivalents:

la. 11 bambino stava giocando nel prato e il suo compagno gli lanci6 la pallache egli riusci ad afferrare al volo, dopo una lunga corsa (uccello -ospite).

la. The child was playing in the garden and his mate threw him the ball thathe was able to catch on the flight, after a long run (bird - guest).

lb. Era cost intelligente ed acuto che non aveva bisogno di lunghe

187

188 COLOMBO

spiegazioni, ma riusciva ad aqerrare al volo le parti essenziali di unproblema (idea - luce).

lb. He was so clever and sharp that he did not need long explanations, butcould catch on the flight (to grasp immediately) the essential parts of theproblem (idea - light).

2a. Per appendere un quadro, bisogna scegliere it punto sul muro in cuibattered it chiodo, e poi dare dei colpi precisi (martello - omaggio).

2a. To hang a picture, one needs to choose the point on the wall where to hitthe nail, and then strike accurately (hammer - present).

2b. Tutti gli amici gli ripetevano continuamente che doveva trovarsi un altrolavoro, e specialmente Luigi, il suo piu caro amico, continuava a battereil chiodo, ma inutilmente (insistenza - preavviso).

2b. All his friends kept telling him that he ought to find another job, inparticular Luigi, his dearest friend, kept hitting the nail, (to insist) withno result (insistence - notice).

3a. 11 pacco era confezionato cos'i accuratamente che dovette prendere laforbice per tagliare la corda che lo legava strettamente (spago - tasto).

3a. The parcel was so accurately prepared that he had to use the scissors tocut the rope that tied it so tightly (string - key).

3b. 11 prigioniero pag6 una somma enorme al guardian della cella e riusci atagliare la corda, dopo molten peripezie (fuga - onda).

3b. The prisoner payed the warden a huge sum and succeeded in cutting therope, (to escape) after many perils (flight - wave).

4a. Dopo aver dipinto l'inferriata con la vernice l'operaio aveva lasciatocadere delle macchie di colore clappertutto, e si trovo il pollice verde,perci6 se lo dovette lavare (rosso - tardo).

4a. After he had painted the railing with the varnish the workman spilled thepainting everywhere, and he discovered that his thumb was green, so hehad to wash it (red - slow).

4b. 11 balcone pieno di bellissimi fiori delta casa di fronte e curato da unasignora con il pollice verde, che conosco bene (piante - motivo).

4b. The balcony of the house opposite ours, which is full of beautifulflowers, is taken care of by a lady with the green thumb, (to have agreen finger) whom I know well (plants - motive).

5a. 11 giorno del suo compleanno tutti i suoi amici per festeggiarlo volevanotirargli gli orecchi (occhi - acqua).


5a. On his birthday all his friends wanted to pull his ears to celebrate (eyes -water).

5b. La mamma si alz6, lo guard severamente e gli disse che doveva tirargligli orecchi e lo mond6 a letto senza cena (castigo - postale).

5b. His mother stood up, looked at him severely and told him that she oughtto pull his ears, and sent him to bed without supper (punishment -postal)

6a. Quella palestra e molto attrezzata ma c'6 poco spazio, I'altro giorno,mentre sollevavo un peso, urtai contro lo spigolo nell'alzare il gomito, emi feci molto male (braccio - denaro).

6a. That gymnasium is very well equipped but there is not enough space,the other day while lifting weights, I bumped against the edge of acorner in raising the elbow and I hurt myself (arm - money).

6b. All'uscita dell'osteria ho visto il vecchio Giovanni che barcollava estava per cadere, perched si sa che gli piace alzare il gomito ogni tanto(vino - arma).

6b. Coming out of the pub I saw old John who was staggering and wasgoing to fall, as it is well known that he likes to raise his elbow (to drinktoo much) every now and then (wine - weapon).

7a. A causa dell'incidente nell'autostrada si era prodotto un enorme ingorgodi traffico, ed era impossibile farsi strada, perci6 siamo rimasti fermiparecchie ore (citta - voce).

7a. Because of the accident in the motorway a huge traffic jam had formed,and it was impossible to make one's way, therefore we had to stop forseveral hours (town - voice).

7b. Non perde alcuna occasion per mettersi in mostra, quell'arrivista delmio collega cerca sempre di avere successo e vuole ad ogni costofarsistrada, anche in modo disonesto (ambizione - paesaggio).

7b. He doesn't miss any opportunity to make an exhibition of himself, thatambitious colleague of mine is always trying to be successful and wantsto make his way at any cost, even dishonestly (ambition - landscape).

8a. Nel teatrino, i bambini si divertivano molto a vedere i buffi movimentidelle marionette, sospese ad un filo, e manovrate dall'abile artista (lana- data).

8a. In the puppet theater, the children enjoyed themselves looking at thefunny movements of the puppets, hanging from a thread, and moved bythe skilled artist (wool - date).

189

190 COLOMBO

8b. L'esploratore durante il viaggio fu costretto ad affrontare numerose pe-ripezie, e molte volte la sua vita fu sospesa ad unfilo ma riusc! sempre acavarsela (rischio - squadra).

8b. During the journey the explorer was compelled to face many perils, andmany times his life hung by a thread but he managed to get out of theproblems (risk - team).

9a. Dopo l'operazione Luigi si era molto indebolito e non riusciva neppure acamminare, fu quindi molto contento quando lentamente riusci a fare ilprimo passo, e a ristabilirsi un po' (piede - conto).

9a. After the operation Luigi was very weak and couldn't even walk, there-fore he was very happy when he slowly managed to take the first step,and feel better (foot - bill).

9b. Era da tempo che volevo conoscere quella ragazza cosl timida, quandola incontrai alla festa sapevo che dovevo fare il primo passo, e parlarle(approccio - ammasso).

9b. For a long time I've been wanting to know that girl so shy, when I mether at the party I knew I was supposed to take the first step and talk toher (approach - bulk).

IOa. La medicina era cosl amara che il bambino non voleva ingoiarla, e glisembrb di sputare veleno, da quanto cattiva era (mortale - cammino).

IOa. The medicine tasted so bitter that the child refused to swallow it, and itseemed like he was spitting poison, because it tasted so bad (mortal -walk).

lOb. 11 successo e la bellezza della sua amica le facevano cosl rabbia chequando ne parlava sembrava sputare veleno, da quanto male ne diceva(invidia - agenzia).

I Ob. The success and the beauty of her friend made her so furious that whenshe talked about it it looked like she was spitting venom, so ill did shespeak of her friend.

I la. D'estate siamo stati all'aperto di notte a guardare il cielo, e mi ha fattovedere le stelle, e insegnato i loro nomi (luna - gara).

11 a. In the summer we used to stay outside during the night to watch the sky,and he has made me see the stars, and taught their names (moon -game).

I lb. Nell'autobus affollato, un signore mi ha pestato e mi ha fatto vedere lestelle, e allora l'ho insultato (dolore - affare).

I lb. In the crowded bus a man stepped on my toes and made me see the stars(to see the stars by daylight) so I insulted him (pain - business).


12a. A giudicare dai numerosi buchi nel vecchio mobile in cucina, qualcosalo rode, forse un tarlo (denti - riva).

12a. Judging from the many holes in the old kitchen cupboard, something isgnawing it, perhaps a wood-worm (teeth - shore).

12b. Giovanni mi sembra molto teso e preoccupato negli ultimi giorni,qualcosa lo rode, ma non so capire cosa (angoscia - assalto).

I2b. Giovanni has been looking very tense and worried in the past few days,something is gnawing him, (to be worried) I do not know what (anguish- assault).

13a. La maestra a lezione spieg6 che i serpenti e i rettili sono animali asangue freddo (caldo - ricco).

13a. During the lecture the teacher said that snakes and reptiles are coldblooded animals (warm - rich).

13b. La situazione era molto complessa e richiedeva che lui avesse auto-controllo e agisse a sanguefreddo, per poterla risolvere (calma - cam-bio).

13b. The situation was very complicated and required him to have self-control and to act in cold blood, to be able to get through (calm -change).

14a. Durante quella terribile estate dovette rimanere in citta, e l'unico modoper sopportare il caldo era una doccia fredda, piui volte al giorno (brividi- cresta).

14a. During that awful summer he had to stay in town, the only way to standthe heat was a cold shower, many times during the day (shiver - crest).

14b. La notizia imprevista arrivata ieri e stata una doccia fredda, e mi harovinato la giornata delusione - perfezione).

14b. The unexpected yesterday news was a cold shower, (nasty surprise) itwasted my day (disappointment - perfection).

15a. 11 clown per far ridere i bambini si fece portare una torta di pannamontata, in cui voleva ficcare il naso, per poi lanciarla via (bocca -libro).

15a. To make children laugh the clown had a cream cake brought in, wherehe wanted to thrust his nose, and that he wanted afterward throw away(mouth - book).

15b. 11 tuo amico mi ha fatto molte domande, mi sembra un tipo cheficca itnaso anche dove non dovrebbe (curioso - affetto).

191

192 COLOMBO

15b. Your friend asked me many questions, he looks like someone whothrusts his nose everywhere, even where he should not (curious - affec-tion).

16a. Per preparare una buona granatina, si deve prendere l'apposito arneseper rompere il ghiaccio, altrimenti non viene bene (neve - pelo).

16a. To prepare a good granatina, one must take the right tool to break theice, otherwise it doesn't come out properly (snow - hair).

16b. 11 signore seduto nel treno si annoiava e si chiedeva come riuscire a farconversazione con il tizio seduto di fronte, ma non sapeva come fare arompere il ghiaccio (iniziativa - testimone).

16b. The man who was sitting in the train was getting bored and was wonder-ing how to start conversation with the guy sitting in front of him, butcould not break the ice (initial ive - witness).

Materials used in Experiment 2 (a - literal contexts, b - ambiguous contexts);in parentheses are the control words, in italics the English translation of thefigurative meaning of the idiom, or the English equivalent:

la. Per fare una granatina doveva rompere (spaccare) il ghiaccio che si eraformato nel frigo.

la. To make a granatina he had to break (crack) the ice that had formed inthe fridge.

lb. Attratto dalla ragazza l'uomo cercava timidamente di rompere (spac-care) il ghiaccio (to break the ice) che si era formato sul vetro.

lb. Attracted by the girl the man was shyly trying to break (crack) the icethat had formed on the glass.

2a. All'arrivo della scatola prese la forbice per tagliare (rompere) la cordache legava il pacco.

2a. As the box arrived he took the scissors to cut (break) the rope that tiedthe parcel.

2b. Per il prigioniero era indubbiarnente molto difficile tagliare (rompere) lacorda (to escape) che legava il pacco.

2b. Undoubtedly it was very difficult for the prisoner to cut (break) the ropethat tied the parcel.

3a. Era nella panna che il pagliaccico per far ridere ficcava (poneva) il nasoche era bello rosso.

3a. It was in the cream that the clown, to make people laugh, thrusted (put)his nose that was very red.

3b. Ero stupefatta quando mi sono accorta dove ficcava (poneva) il naso (tothrust one's nose) che era bello rosso.


3b. I was astonished when I realized where he thrusted (put) his nose thatwas very red.

4a. Mentre faceva ii bagno Giovanni voleva gettare (strizzare) la spugna chespruzzava dappertutto.

4a. While he was taking a bath Giovanni wanted to throw (squeeze) thesponge that was splashing everywhere.

4b. Ormai stanco a quel punto Giovanni voleva gettare (strizzare) la spugna(to give up) che spruzzava dappertutto.

4b. Now weary, at that point Giovanni wanted to throw (squeeze) thesponge that was splashing everywhere.

5a. Manovrata dall'abile artista la marionette era sospesa ad (tenuta da) unfilo che pendeva dal soffitto.

5a. Maneuvered by the skilled artist the puppet was suspended to (kept by) athread that was hanging from the ceiling.

Sb. Per tutto il tempo si era sentito sospeso ad (tenuto da) un filo (to hang bya thread) che pendeva dal soffitto.

Sb. All time long he had felt suspended to (held by) a thread that hung fromthe ceiling.

6a. Talvolta osservando il cielo capita di vedere (guardare) le stelle cheformano l'Orsa Maggiore.

6a. Sometimes watching the sky one happens to see (look at) the stars thatform the Great Bear constellation.

6b. Girandosi improvvisamente quell'uomo mi ha fatto vedere (guardare) lestelle (to see stars by daylight) che formano l'Orsa Maggiore.

6b. Turning suddenly the man has made me see the stars that form the GreatBear constellation.

7a. 11 falegname prese il pezzo di legno e si mise a battere (colpire) il chiodoche fisso al muro.

7a. The carpenter took the piece of wood and started hitting the nail that hefastened to the wall.

7b. Continuava a battere (colpire) il chiodo (to insist) che doveva fissare almuro.

7b. He kept hitting (striking) the nail that he had fastened to the wall.

8a. 11 cavallo era troppo fiacco per tirare (spingere) la carretta che era pienadi mattoni.

8a. The horse was too weak to pull (push) the cart that was full of bricks.

194 COLOMBO

8b. La vecchia era sfinita e non ne poteva piji di tirare (spingere) la carretta(to drudge) che era piena di mattoni.

8b. The old woman was worn out and done up with pulling (pushing) thecart that was full of bricks.

9a. 11 medico chiese al paziente di stringere (far vedere) i denti che eranosottoposti alla cura.

9a. The doctor asked the patient to clench (let see) the teeth that were undertreatment.

9b. Ai soldati fu raccomandato di stringer (far vedere) i denti (to enduresomething) che erano sottoposti alla cura.

9b. The soldiers were advised to clench (let see) the teeth that were undertreatment.

IOa. A causa di tutte le spese fatte voleva fare (vedere) i conti che si eranoaccumulate.

IOa. Because of all the expenses that had been made he wanted to do (see) theaccounts that had piled up.

IOb. Con un cenno mi ha fatto capire di volere fare (vedere) i conti (toreckon) che si erano accumulate

lOb. With a sign he made me understand he wanted to do (see) the accountsthat had piled up.

I la. Mi disse che aveva visto il gatto in cucina mangiarsi (ingoiare) il fegato(to be worried) che era la sua passione.

I la. He told me he had seen the cat in the kitchen eating (ingesting) the liverthat he was very fond of.

I lb. Lo aveva visto come si mangiava (ingoiava) il fegato (to be worried) cheera la sua passione.

I lb. He had watched how (he/she/it/) ate (ingested) the liver that he/she/itwas very fond of.

12a. Mentre dipingeva si guard le mani e si accorse di avere il pollice verde(rosso) che pull sullo straccio.

12a. While he was painting he looked at his hands and realized he had thegreen (red) thumb that he cleaned with the rag.

12b. Guardando il suo bellissimo terrazzo aveva scoperto di avere il policeverde, (to have a green finger) che puli sullo straccio.

12b. Looking at his beautiful balcony he had discovered he had the greenthumb that he cleaned with the rag.


13a. II dottore per visitarlo gli ha chiesto di alzare (sollevare) il gomito che siera fratturato recentemente.

13a. To visit him the doctor asked him to raise (lift) the elbow that he hadrecently broken.

13b. L'altro giomo all'osteria con gli amici gli era capitato di alzare (sol-levare) il gomito (to drink too much) che si era fratturato recentemente.

13b. A few days ago while he was at the pub with his friends he happened toraise (lift) the elbow that he had recently broken.

14a. Stefano era affamato e la mamma gli ha dato (offerto) del salame che glie indigesto.

14a. Stefano was hungry and his mother gave (offered) him some salami thathe cannot digest.

14b. t un tipo molto suscettibile e si e offeso perch Giorgio gli ha dato(offerto) del salame (to call someone an ass) che gli e indigesto.

14b. He is a very susceptible guy and he got offended because Giorgio hasgiven (offered) him some salami that he cannot digest.

15a. 11 cacciatore per prendere la mira ha dovuto chiudere (coprirsi) un oc-chio che gli faceva male.

15a. To take aim the hunter had to close (to cover) an eye that ached.15b. Per quella faccenda ha dovuto chiudere (coprirsi) un occhio (to over-

look) che gli faceva male.15b. Because of that affair he had to close (to cover) an eye that ached.

16a. Durante un brusco movimento della nave il capitano ha perso (rotto) labussola che serve per navigare.

16a. During a rough movement of the ship the captain has lost (broken) thecompass that is used for navigation.

16b. Dopo l'incidente Stefano ha perso (rotto) la bussola (to lose one's head)che gli avevo regalato.

16b. After the incident Stefano has lost (broken) the compass that I had givenas a present to him.

Materials used in Experiment 3 (a - literal contexts, b - figurative contexts); inparentheses are the words for the control condition, in italics the English transla-tion of the figurative meaning or the equivalent English idiom:

la. Per fare una granatina doveva prendere l'apposito arnese per rompere ilghiaccio che era in frigo.

196 COLOMBO

1 a. To prepare a granatina he had to take the appropriate device to break theice that was in the fridge.

lb. Dopo il litigio con la fidanzata Mario voleva parlarle ma non riusciva arompere il ghiaccio (break the ice) che era in frigo (il piatto che era infrigo).

lb. After the row with her girlfriend Mario wanted to talk to her but couldnot break the ice (plate) that was in the fridge.

2a. All'arrivo della scatola l'uomo prese velocemente la forbice per tagliarela corda che legava ii pacco.

2a. When the box arrived the man hurried to take the scissors to cut the ropethat tied the parcel.

2b. Nonostante le guardie che gli sparavano il prigioniero era riuscito atagliare la corda (to escape) che legava il pacco (il nastro che legava ilpacco).

2b. In spite of the guards that fired at him the prisoner managed to cut therope (ribbon) that tied the parcel.

3a. Era nella torta di panna che il pagliaccio per far rider ficcava il naso cheera molto lungo.

3a. It was in the cream cake that the clown, to make people laugh, thrust hisnose that was very long.

3b. Quella donna voleva sempre sapere tutto di tutti e mi ha irritate scopriredove ficcava il naso (to thrust one's nose) che era molto lungo (il ditoche era molto lungo.

3b. That woman always wanted to know everything of everyone and itannoyed me to find out where she thrusted her nose (finger) that wasvery long.

4a. Nella vasca dove faceva il bagno il bambino voleva gettare la spugna cheperdeva acqua.

4a. In the bathtub where he was bathing the child wanted to throw thesponge that spilled water.

4b. Deluso l'atleta aveva capito di non farcela a vincere e aveva deciso digettare la spugna (to give up) che perdeva acqua (il secchio che perdevaacqua).

4b. The disappointed athlete had understood that he could not make it andhad decided to throw the sponge (pail) that spilled water.

5a. Manovrata dall'abile artista la marionetta nel teatrino era sospesa ad unfilo che pendeva dal soffitto.


5a. Maneuvered by the skilled artist the puppet in the little theater wassuspended to a thread that hung from the ceiling.

5b. Dopo il terribile incident la vita dell'esploratore era stata sospesa ad unfilo (to hang by a thread) che pendeva dal soffitto (il cavo che pendevadal soffitto).

5b. After the accident the explorer's life had been suspended to a thread(cable) that hung from the ceiling.

6a. Andrea una sera ha portato Maria in giardino e mostrando il cielo le hafatto vedere le stelle che formano l'Orsa Maggiore.

6a. One evening Andrea took Maria in the garden and pointing to the sky hemade her see the stars that form the Great Bear constellation.

6b. Quell'uomo in autobus mi ha dato un pestone cosi forte e mi ha fattovedere le stelle (to see stars by daylight) che formano l'Orsa Maggiore(gli astri che formano l'Orsa Maggiore).

6b. The man in the bus stepped in my toes so heavily, and he made me seethe stars (luminaries) that form the Great Bear constellation.

7a. 11 falegname ha preso il pezzo di legno e si e messo a battere il chiodoche doveva fissare al muro.

7a. The carpenter took the piece of wood and started hitting the nail that hehad to fasten to the wall.

7b. Quando vuole una cosa Sandro la chiede insistentemente e finches nonl'ha ottenuta continua a battere il chiodo che deve fissare al muro (illegno che deve fissare al muro.)

7b. When he wants something, Sandro asks with insistence for it and, untilhe has obtained it, he keeps hitting the nail (wood) (to insist) that hemust fasten to the wall.

8a. 11 cavallo e troppo magro per trascinare un grosso peso e non ce la fa atirare la carretta che e mezza rotta.

8a. The horse is too thin to drag a heavy weight and cannot manage to pullthe cart that is half broken.

8b. Quando gli ho chiesto se ce la fa a mantenere tutti i suoi figli ha rispostoche e duro tirare la carretta (to drudge) che e mezza rotta (la canna che emezza rotta).

8b. When I asked him if he can manage to support all his children he toldme that it is hard to pull the cart (cane) that is half broken.

9a. L'infermiere chiese al paziente che doveva provare l'apparecchio distringer i denti che gli facevano male.

197

198 COLOMBO

9a. The nurse asked the patient that had to try the dental apparatus to clenchhis teeth that ached.

9b. 11 soldato aveva paura ma ii cornandante aveva raccomandato di andareavanti e di stringer i denti (to endure something) che gli facevano male(i polsi che gli facevano male),

9b. The soldier was scared but the commander had advised to go ahead andto clench his teeth (pulse) that ached.

lOa. Mi disse che aveva visto it gatto in cucina che si era arrampicato sultavolo e si stava mangiando ii fegato che gli piaceva molto.

IOa. He told me that he had seen the cat in the kitchen that had climbed on thetable and was eating the liver that he liked much.

IOb. Per aver perso troppi soldi in borsa lFazionista era disperato e si stavamangiando il fegato (to be worried) che gli piaceva molto (il gelato chegli piaceva molto).

IOb. Having lost too much money at the stock exchange market the share-holder was desperate and was eating the liver (ice cream) that he likedmuch.

I la. Martino pasticciando con i colori si era sporcato e guardandosi le maniaveva visto che aveva il pollice verde che pull sullo straccio.

I la. Martino playing with the painting colours had got dirty and looking athis hands had seen that he had the green thumb that he cleaned on therag.

I lb. Vedendo le rigogliose piante che erano cresciute in casa sua la amiche ledicevano che aveva il pollice verde (to have a green finger) che pullsullo straccio (rosso che pull sullo straccio).

1 lb. Looking at the luxuriant plants that had grown in her house her friendstold her that she had the green (red) thumb that she cleaned on the rag.

12a. 11 dottore per visitare l'arto colpito in un incident di lavoro gli chiese dialzare il gomito che era fratturato.

12a. To visit the limb injured in a work accident the doctor asked him to raisethe elbow that had broken.

12b. Tutte le volte che va fuori con gli amici a cenare mangia troppo e poi glisuccede di alzare il gomito che si era fratturato (il braccio che si erafratturato).

12b. Every time he dines out with his friends he eats too much and then hehappens to raise the elbow (arm) (to drink too much) that had broken.


13a. Stefano non aveva mangiato a colazione ed era affamato e la mamma gliha dato del salame che gli piace molto.

13a. Stefano had not eaten for breakfast and he was hungry and his mothergave him some salami that he likes very much.

13b. Stefano e molto suscettibile e si e offeso perch la mamma lo ha sgridatoe gli ha dato del salame (to call someone an ass) che gli piace molto (delgelato che gli piace molto).

13b. Stefano is very susceptible and he took offense because his motherscolded him and gave him some salami (ice cream) that he likes verymuch.

14a. 11 cacciatore ha preso il fucile e per prendere la mira ha dovuto chiudereun occhio che era gonfio.

14a. The hunter took the gun and to take aim he had to close an eye that wasswollen.

14b. Era stato fatto troppo chiasso attorno a quella vicenda e il giudice avevadeciso di chiudere un occhio (to overlook) che era gonfio (il naso che eragonfio).

14b. There had been too much rumor about that affair and the judge haddecided to close an eye (nose) that was swollen.

15a. Durante un brusco movimento della nave il capitano ha perso la bussolache serve per navigare.

15a. During a rough movement of the ship the captain has lost the compassthat is used for navigation.

15b. Da quando si e innamorato di quella ragazza Gino non capisce piuniente, ha perso la bussola (to lose one's head) che serve per navigare (iltimone che serve per navigare).

15b. Since he has fallen in love with that girl Gino does not understandanything anymore, he has lost the compass (rudder) that is used fornavigation.

16a. leri Giulia e andata a passeggiare sulla riva del mare e ha preso ungranchio che camminava sulla sabbia.

16a. Yesterday Giulia went for a walk on the seashore and she took a crabthat was crawling on the sand.

16b. All'inizio Giulia pensava che Giuseppe fosse cotto di lei ma aveva presoun granchio (to take a blunder) che camminava sulla sabbia (una moscache camminava sulla sabbia).

200 COLOMBO

16b. At the beginning Giulia thought that Giuseppe was lovesick for her butshe had taken a crab (fly) that was crawling on the sand.

17a. 11 bruco dopo aver strisciato lentarmente lungo il ramo aveva mangiato lafoglia che era sulla pianta.

17a. Having slowing crawled along the branch the caterpillar had eaten theleaf that was on the plant.

17b. Avevo cercato di tenergli nasccsta quella storia ma lui aveva mangiato lafoglia (to take the hint) che era sulla pianta (I'insetto che era sullapianta).

17b. I had tried to keep that story concealed but he had eaten the leaf (insect)that was on the plant.

18a. 11 medico chiese al paziente che si lamentava di mostrare i denti cheerano guasti.

18a. The doctor asked the patient that was complaining to show the teeth thatwere decayed.

18b. Mario era un tipo molto aggressive e quando occorreva sapeva mostrarei denti (to show one's teeth) che erano guasti (la radio che era guasta).

18b. Mario was a very aggressive guy and when needed he could show theteeth (radio) that were decayed.

Syntactic and SemanticProcessing During IdiomComprehension:Neurolinguistic andPsycholinguistic Dissociations

Robert R. PetersonIndiana University

Curt BurgessUniversity of California - Riverside

Research in psycholinguistics traditionally has been guided by the assumptionthat the comprehension of language involves the construction of increasinglyabstract levels of representation over time. For example, in understanding anutterance, a listener initially may construct fairly shallow descriptions of theutterance's acoustic and phonetic form. These shallow representations, then,guide (or serve as input to) intermediate-level processors (e.g., lexical and syn-tactic processors) whose output, in turn, fuels higher level processing systems(e.g., those that determine the meaning of individual sentences, and build repre-sentations of the discourse as a whole). The adoption of this levels-of-processingframework has influenced significantly the kinds of theoretical issues that psy-cholinguists have tended to view as central to the field. Thus, researchers havebeen preoccupied with addressing such issues as the "psychological" reality ofparticular linguistic levels of representation, the temporal relationship amonglevels (i.e., whether levels of processing operate in strict temporal succession, orwhether they operate in cascade), and the ways that a given level responds toambiguous input (e.g., whether it computes a single interpretation or insteadconstructs multiple representations). One issue in particular, however, that hasmotivated an impressive amount of research over the past decade concerns themodular versus interactive nature of the various processing systems. This re-search specifically has sought to determine whether there is only bottom-upcommunication among processing levels (i.e., where information flows onlyfrom lower to higher processors), or whether, in addition, higher level processorssometimes can guide the operation of lower level systems. Thus, it is the natureof the dependencies that exist among processing levels that is at the heart of themodularity debate.

201

202 PETERSON AND BURGESS

In the present chapter, we specifically consider the processing dependenciesthat exist between the syntactic and semantic systems. To investigate these de-pendencies, we have focused on the comprehension of idiomatic phrases (e.g.,kick the bucket). Although these kinds of phrases generally have been ignoredwithin the modularity debate, we argue that research on idiom comprehensionmay provide important insights into the processing relationships that exist be-tween syntax and semantics. Idioms seem particularly well suited for this taskbecause they often exhibit intriguing dissociations between these two levels ofrepresentation. That is, many idioms are structured in syntactically appropriateways, although this structure cannot be used, in combination with the meaningsof the idiom's individual words, to derive the idiom's figurative interpretation.For example, kick the bucket is a well-formed verb phrase, but its idiomaticmeaning bears little relation to the phrase's syntactic structure. This lack ofcorrespondence between levels creates something of a dilemma for the com-prehension system: Whereas the syntactic processor could generate a structuraldescription of an idiom in a straightforward manner, the semantic processor, if itused that description (in combination w ith the words of the idiom), would gener-ate a literal interpretation of the idiom., even in contexts where the figurativeinterpretation was intended. Given this characteristic of idiomatic phrases, animportant empirical question that needs to be addressed is whether (and how) theprocessing system might prevent an unintended literal interpretation from beingcomputed. Two solutions to this problem seem possible, both of which haveimportant implications for how the two processing systems might interact. First,the semantic processor simply could ignore the information being fed to it by thesyntactic processor. If this possibility is correct, then it would suggest that thesemantic processor is not obligatorily responsive to bottom-up syntactic input.Second, the semantic processor actively could signal the syntactic processor tocease ongoing structural analysis. According to this view, the operation of syn-tactic processing is dependent on both bottom-up and top-down (i.e., semantic)influences. Investigating how idioms are processed, therefore, may provide uswith a truly unique vantage point from which we can observe the ongoinginteractions among different linguistic systems.

The remainder of this chapter is organized into two main sections. In the firstsection, we briefly review some of the ways that researchers have thought aboutthe syntactic and semantic character of idiomatic phrases. In addition, we presentthe results from a series of experiments that we recently have conducted thatdirectly address the issue of syntactic/semantic dependencies in processing. Twospecific questions were explored in these studies. First, we investigated whetherthe syntactic processor actually computes the structure of an idiom during com-prehension. Second, we tested whether the semantic processor is compelled toderive the literal interpretation of the idiom, based on that structure. As notedearlier, answers to these questions can provide insight into the nature of theinteractions that occur between the syntactic and semantic systems. In the second

9. SYNTACTIC AND SEMANTIC PROCESSING OF IDIOMS

section of the chapter, then, we review some of the neurolinguistic literature onidiom comprehension. This research is particularly relevant to the issues beingraised here because it has shown that there is, in fact, a neural dissociationbetween syntactic processing and the construction of figurative meaning. Specif-ically, it appears that, although syntactic processing tends to be a left hemispherefunction, the comprehension of idiomatic phrases (and other kinds of figurativephrases as well) appears to be a specialization of the right hemisphere. Neuralresearch, therefore, may give us important insights into how the comprehensionsystem utilizes hemispheric asymmetries to deal with dissociations among pro-cessing levels. At the end of this second section we attempt to relate our experi-mental findings to neurolinguistic data, and suggest some experiments that wouldhelp to clarify the role that the two hemispheres play in idiom comprehensionspecifically, and in the interdependencies among levels of processing more gen-erally.

IDIOMS, SYNTACTIC PROCESSING,AND LITERAL MEANING

There are a number of studies in the literature that have investigated the role ofliteral processing during the comprehension of idiomatic phrases. Unfortunately,these studies have not attempted to distinguish between syntactic and semanticanalyses, and therefore they do not provide direct evidence regarding the natureof the dependencies that exist between these levels. Nevertheless, many of themodels that have evolved in the idiom-processing literature make some specificclaims that are (at least indirectly) relevant to the issue we are addressing in thischapter. First, we briefly describe some of these existing models, focusing spe-cifically on the kinds of implications that they might have for the issue ofsyntactic/semantic dependencies. During this discussion, we develop three newmodels that more directly specify the relationship between syntactic and seman-tic processing during the comprehension of idiomatic phrases. We refer to thesenew models as the syntactic dominance, the semantic dominance, and syntactic-semantic autonomy models, respectively. Later in the chapter, we present a seriesof experiments that attempt to distinguish among these three candidates.

Swinney and Cutler (1979) argued that idioms are stored as single entries inthe mental lexicon, and are retrieved in the same manner as any other lexical item(they refer to this view as the lexical representation hypothesis). Thus, accordingto this view, the comprehension of an idiom does not require that a compositionalanalysis be undertaken. Instead, the figurative meaning of an idiom can beretrieved directly (and holistically) from the lexicon. Swinney and Cutler as-sumed that the retrieval of an idiom's figurative meaning occurs in parallel withthe computation of its literal meaning. Figurative processing, however, will tendto conclude more quickly than literal processing because the direct access of the

203


figurative meaning is simpler than the compositional analysis required to derive aliteral interpretation. In support of this claim, Swinney and Cutler demonstratedthat subjects can make a judgment that an idiomatic phrase (e.g., see the light) ismeaningful more quickly than they can make the same judgment about a phrasehaving only a literal interpretation (e.g., get the light).

Several researchers have argued that parallel access occurs not only for idiomspresented in isolation, but also for idioms presented in contexts that clearly biasthe idiom's interpretation (Estill & Kemper, 1982; Ortony, Shallert, Reynolds, &Antos, 1978). This claim generally has been based on demonstrations, likeSwinney and Cutler's (1979), that idioms are no harder, and perhaps are easier toprocess than are literal phrases. The claim that the literal meaning of an idiom isretrieved even in contexts where it is inappropriate implies that literal processingis, in fact, obligatory. In terms of the issue of interlevel dependencies in process-ing, this view suggests that the semantic processor cannot ignore bottom-up inputfrom the parser. That is, even in figuratively biasing contexts, the semanticprocessor apparently is compelled to assemble a literal interpretation of an idi-omatic phrase based (in part) on the phrase's syntactic structure. For the purposesof future reference, we refer to this particular view of syntactic and semanticprocessing as the syntactic dominance model, reflecting the fact that output fromthe syntactic processor automatically guides the operation of the semantic pro-cessor; that is, the semantic processor is obligatorily responsive to structuralinput. Later, we contrast this model with two other possible models of syntactic/semantic dependencies.

In contrast to the Swinney and Cutter (1979) position, Gibbs (1986) arguedthat faster responses to idiomatic than to literal phrases could be accounted forby a serial rather than a parallel processing model. Specifically, he suggestedthat the meaning of an idiom always precedes literal processing, with the literalmeaning being derived only if the figurative meaning is found to be anomalousgiven the immediate context. To provide evidence for this hypothesis, Gibbspresented subjects with short stories that ended with a sentence containing anidiomatic expression. The story was biased for either a literal or a figurativeinterpretation of the idiom. After reading the story, subjects were presented witha target sentence that paraphrased either the literal or figurative meaning of theidiom, or was unrelated to the idiom's meaning. The subjects' task was to decideif the target sentence was meaningful. Gibbs found that responses to figurativeparaphrases were facilitated following literal contexts, thus suggesting that theidiomatic meaning had been retrieved even though it was contextually inappro-priate. On the other hand, responses to literal paraphrases were not facilitatedfollowing figuratively biased contexts, suggesting that the literal meaning of theidioms had not been computed. Gibbs argued that these data demonstrate that thefigurative meaning of an idiom is always computed, whereas the literal meaningis computed only if the figurative meaning is inappropriate.

An important implication of the Gibbs (1986) model is that the semantic


processor can terminate literal processing of an idiomatic string. Although thisclaim is intriguing, there are a number of important questions regarding thetermination process that are left unspecified in the model. For example, whereasGibbs seems to imply that the figurative meanings of idioms can be retrievedwithout any analysis of their literal meanings, it is never made clear how thesemantic system actually decides to abort literal processing. Of particular diffi-culty for the model is the fact that (literal) semantic interpretation appears tooccur in rapid fashion during on-line comprehension, presumably proceedingon a word-by-word basis (Crain & Steedman, 1985; Marslen-Wilson, 1985;Marslen-Wilson & Tyler, 1980). Thus, in order for there not to be any literalinterpretation of an idiomatic string, the termination of literal processing mustoccur immediately upon the presentation of the first word of the idiom (i.e., iftermination did not occur immediately, then, contrary to Gibbs' claim, the firstword of the idiom would be integrated rapidly, and literally, with the precedingtext). It seems doubtful, however, that immediate termination could occur in allcontexts. Consider, for example, the sentence "The man kicked the bucket."According to Gibbs' model, a listener, upon hearing the word kicked wouldterminate literal analysis of the sentence (i.e., would not integrate kicked with theman) and therefore would interpret the remainder of the verb phrase figuratively.The problem with this scenario is that there is no cue in the context to indicatethat the rest of the sentence will continue idiomatically. Indeed, our intuition isthat a sentence beginning with "The man kicked the" will turn out, more oftenthan not, to be a literal statement (e.g., 'The man kicked the ball"), rather than afigurative assertion (e.g., "The man kicked the bucket"). Of course, it may bethat the system suspends literal analysis any time a word is encountered thatcould be part of an idiomatic phrase. Such a strategy, however, would be terriblyinefficient. Given that most words in the language can be part of some idiomaticphrase, this strategy would result in routine suspension of literal analysis, andhence, in tremendous delays in interpretation. As noted previously there is ampleevidence in the sentence processing literature indicating that such delays inprocessing typically do not occur.

Thus, we disagree with Gibbs' (1986) claim that literal analysis always mustawait the evaluation of possible figurative interpretations of a given phrase. It isimportant to note, however, that we are not claiming that a full literal interpreta-tion of an idiomatic phrase must be constructed in all contexts (as the syntacticdominance model proposes). That is, it seems logically possible that, under somecircumstances, literal analysis, in fact, could be suspended. In particular, thispotentially could occur when when an idiom is presented in a figuratively biasingcontext. For example, consider the sentence fragment, "The man was old andfeeble and it was believed that he would soon kick...." Intuitively, it seemsthat, in this context, one can recognize kick as being the beginning of the idiomkick the bucket. If the processing system in fact can retrieve the idiom at thispoint, then the successful integration of the meaning "die" into the developing

205


representation of the sentence could signal the semantic processor to disregardthe literal implication of the remainder of the phrase. Thus, if the sentencefinished with the words the bucket, the meaning of those words, and theirgrammatical function, might not be considered by the semantic processor. Cac-ciari and Tabossi (1988) recently provided evidence consistent with this scenario.These researchers demonstrated that when subjects are biased to interpret thebeginning of a phrase idiomatically, the final words of the phrase are not in-terpreted literally (or more specifically they showed, using a cross-modal prim-ing procedure, that the final word of the idiom does not prime its literal associ-ates).

Both Gibbs' (1986) and Cacciari and Tabossi's (1988) hypotheses contradictthe claims of syntactic dominance model (as outlined earlier), because eachproposes that the computation of literal meaning can be terminated. An importantquestion remains, however, regarding the actual scope of this termination pro-cess. In particular, it is not clear which levels of processing (other than onecomputing the semantic structure of the sentence) might be aborted. Can theparser, for example, terminate its structural analysis of an idiomatic string oncethe figurative meaning of the idiom has been retrieved? We briefly describe twopossible models that provide different answers to this question (we refer to thesemodels as the semantic dominance mode] and the syntactic-semantic autonomymodel, respectively). The models differ from one another specifically in terms ofthe type of top-down dependencies that they posit between semantics and syntax.Because each model posits that literal interpretation can be disabled, they bothcontradict the claims made by the syntactic dominance model.

According to the semantic dominance model, the semantic processor, uponretrieving the idiomatic meaning of an idiom, will integrate that meaning withthe representation of the preceding text, and will signal the syntactic processor todiscontinue further structural analyses. This model predicts, therefore, that nei-ther a full structural description of an idiom, nor its literal interpretation, will bederived necessarily in a figuratively biasing context. Compared to the syntacticdominance model, the semantic processor in this model is more independent ofsyntactic processing because it does not interpret structural input automatically.In addition, in this model the semantic processor exerts more control over thesyntactic processor because the establishment of a figurative interpretation of theidiom can inhibit further structural analyses. Such top-down influences makethe model particularly compatible with interactive models of sentence process-ing (e.g., Altman & Steedman, 1988; Crain & Steedman, 1985; Taraban &McClelland, 1988).

The final model that we consider is similar to the semantic dominance modelin that it proposes that the semantic processor can terminate its analysis of theliteral meaning of an idiomatic phrase. However, in this model there is nofeedback from the semantic to the syntactic processor, hence the termination ofthe literal analysis does not inhibit further syntactic processing (as it does in the


TABLE 9.1Sample Sentence Primes and Targets from Experiments 1, 2, and 3

Targets

Exp. 1 Exp. 2 Exp. 3

Sentence Primes Verb Noun "Verb ' "Noun" Abstract Concrete

Idiomatic ContextsThe man was old grow town glatted glatter truth shelf

and feeble and itwas believed thathe would soonkick the

The business was lend shirt fitted hitter guilt pearlrunning quitesmoothly so themanager warnedthe disgruntledsecretary not torock the

Literal ContextThe soccer player grow town glatted glatter truth shelf

slipped whentried to kick the

The only way the lend shirt ditted hitter guilt pearlmother could gether baby to go tosleep was to layhim down andgently rock the

semantic dominance model). We refer to this model as the syntactic-semanticautonomy model. This model predicts, therefore, that a full syntactic descriptionof an idiom always will be derived, even though the idiom's full literal meaningwill not necessarily be computed. In comparison to both the syntactic dominanceand the semantic dominance models, this final model assumes the greatest degreeof independence between the two levels of processing, because neither processoris obligatory responsive to the other's output. Of the models that we have consid-ered, this model is the most compatible with modular models of sentence pro-cessing (Ferreira & Clifton, 1986; Rayner, Carlson, & Frazier, 1983).

Deciding which of the aforementioned three models (i.e., the syntactic domi-nance, the semantic dominance, and the syntactic-semantic autonomy models)best describes the nature of syntactic/semantic dependencies in on-line com-prehension requires answering two specific questions. First, it must be deter-mined whether the syntactic structure of an idiomatic phrase is necessarily com-

207


puted in a figuratively biasing context. The syntactic dominance and thesyntactic-semantic autonomy models propose that a full description is derived,whereas the semantic dominance model suggests that syntactic analysis is termi-nated during the course of processing an idiomatic phrase. The second criticalquestion, then, is whether the full literal meaning of an idiom is computedobligatorily. The syntactic dominance model claims that it is, whereas the othermodels suggest that termination of literal analysis is possible. Overall, then, thethree models make unique predictions with respect to these two questions (seeTable 9.1 for a summary of these predictions), and thus they lend themselves todirect empirical test. In the next section, we briefly describe the results of a set ofthree experiments that we have conducted to distinguish among the models. Inthe first two experiments, we exploded the syntactic processing of idiomaticphrases. In the final experiment, we tested whether the literal meaning of anidiom is computed in contexts that bias a figurative interpretation.

Syntactic Processing of Idiomatic Phrases

The nature of syntactic processing has received some attention in the idiomcomprehension literature. For example, Gibbs and Gonzalez (1985) investigatedhow the processing of an idiom varies as a function of the idiom's syntacticfrozenness (see also Reagan, 1987; Swinney & Cutler, 1979). Frozenness refersto the extent to which an idiom can undergo syntactic transformations and stillretain its figurative meaning. Some idioins are extremely frozen and can undergofew transformations (e.g., "John kicked the bucket" can mean "John died,"whereas "The bucket was kicked by John" cannot), whereas others are moreflexible (or unfrozen), and can undergo several different transformations (e.g.,both "Her father laid down the law' and "The law was layed down by herfather" seem to mean that "Her father gave strict orders"). In addition to theseextremes, idioms also can be intermediate in their level of syntactic flexibility(retaining their meaning under some, but not all, transformations). Thus frozen-ness appears to be a continuous rather than a strictly dichotomous dimension(Fraser, 1970; Gibbs & Gonzalez, 1985; Reagan, 1987).

Gibbs and Gonzalez (1985) argued that the more frozen an idiom is, the easierit is to process. This processing advantage might arise because, with increasinglevels of frozenness, less syntactic analysis is required in order to derive themeaning of the idiom. This view of frozenness would seem to suggest that thethree models that we have proposed might be too extreme in their views ofsyntactic processing. Rather than the all-or-none view taken by these models(i.e., the models propose that a full syntactic description either is, or is not,derived), it might be better to think of syntactic processing in more continuousterms, with idioms undergoing different degrees of structural analysis dependingon their level of frozenness. It is important to note, however, that the discreteview that we have adopted is not necessarily incompatible with the continuous

9. SYNTACTIC AND SEMANTIC PROCESSING OF IDIOMS 209

view outlined by Gibbs and Gonzalez. The apparent differences between theviews arise because they focus on slightly different questions. In our models, weare interested in whether the syntactic processor automatically computes thestructure of an idiomatic phrase, but we do not necessarily address the questionof whether a full syntactic description is required to access an idiom's figurativemeaning. In contrast, it is this latter question that is of specific interest to Gibbs.These two questions, however, are logically independent. For example, it mightbe that the syntactic processor obligatorily performs a complete structural analy-sis of an idiom (consistent with the syntactic-semantic autonomy model), eventhough idioms differ in the amount of structural analysis that is required toretrieve their figurative meanings. We return to this issue following the presenta-tion of Experiments I and 2, and also address it in the second section of thischapter when we consider the neurolinguistic underpinnings of idiom com-prehension.

To investigate the syntactic processing of idiomatic phrases, we (Peterson,Burgess, Dell, & Eberhard, 1989) used a sentence-priming procedure. Subjectswere presented with auditory sentence primes that ended with incomplete idioms(e.g., kick the . . . ). Sentences were biased for either a literal (ball) or idiomatic(bucket) noun completion. I Examples of each of these sentence types are givenin Table 9.2. Immediately following the final word of each incomplete sentence,a visual target word was presented on a computer screen. Subjects were in-structed to name the target word as quickly as possible, and their naming laten-cies were recorded. The target words were always semantically unrelated to thecontexts. Critically, however, the syntactic appropriateness of the target wordwas manipulated. Specifically, on a given trial the target was either a noun (anappropriate completion) or a verb (an inappropriate completion; see Table 9.2 forexamples of these targets). A number of studies in the literature have demon-strated that responses are faster (in both lexical decision and naming) to syntac-tically appropriate than to inappropriate targets (e.g., Goodman, McClelland, &Gibbs, 1981; West & Stanovich, 1986; Wright & Garrett, 1984). We used thissyntactic priming effect as an indicator of ongoing structural analysis. We werespecifically interested, then, in comparing the magnitude of priming betweenliterally and figuratively biased sentence primes in order to determine if idi-omatic strings undergo normal syntactic analysis. Thus, we were expecting that,for literally biased sentences, responses would be faster to the noun than the verbtargets.2 At issue, then, was whether a comparable effect occurs with figurativelybiased sentences. If it does, then it suggests that the syntactic structures ofidiomatic strings are derived during on-line comprehension (consistent with both

'Using a Cloze procedure, literal and idiomatic contexts were found to be equally constraining fortheir expected noun completions.

2A pilot study demonstrated that, in isolation, there was no difference in naming latency betweenthe noun and verb targets. Similar results were found for the targets used in Experiments 2 and 3.


TABLE 9.2Empirical Predictions Made by the Three Models

Predictions

Empirical Syntactic Semantic Syntactic-SemanticQuestions Dominance Modei' Dominance Model Autonomy Model

Does an idiom un- YES NO YESdergo a full syn-tactic analysis?

Is the literal mean- YES NO NOing of an idiomcomputed?

the syntactic dominance model and the syntactic-semantic autonomy model). If,however, the priming effect with figuratively biased sentences is eliminated orreduced (compared to the priming effect with literal sentences), it would suggestthat there is a diminution of syntactic processing during the comprehension ofidiomatic phrases (consistent with the semantic dominance model).

The results of Experiment I are shown in Fig. 9.1. As can be seen, verbtargets were named more slowly than were nouns, and this pattern held for bothliteral and idiomatic sentence contexts. Critically, the magnitude of the primingeffect was similar in the two contexts. in fact, the syntactic priming effect wasnumerically larger for the idiomatic sentences than for the literal sentences (29ms vs. 16 ins, respectively), although this interaction was not statistically signifi-cant. These data clearly indicate, then, that there is no diminution of syntacticprocessing when figurative contexts are used. These results are consistent withthe predictions made by the syntactic dominance and the syntactic-semanticautonomy models. Both of these models assume that the syntactic processorobligatorily parses an idiomatic phrase. In contrast, the results of Experiment Ifail to support the semantic dominance model. In that model, parsing of anidiomatic string is discontinued once the phrase has been interpreted figuratively,and therefore the observed syntactic priming effect makes the model less tenable.However, it is premature to reject the model based solely on the results ofExperiment I because the semantic dominance model assumes that parsing isterminated only if the literal meaning cf the idiom no longer is being computed.Thus, disproving the model requires testing for the derivation of literal meaning.We directly address this issue in Experiment 3.

There is a possible criticism of the results from Experiment I that couldqualify our conclusion that idioms undergo full syntactic analysis. Specifically, itis possible that the syntactic priming effect that was obtained in Experiment I isnot a reflection of the syntactic incompatibility of the verb targets per se. Instead,


E 540. -

. 530-

C

.n 520,

CD

E 510-CCCoCo 500.

ACA

Target Type

* VerbEM Noun

iulom Literal

Type of Phrase

FIG. 9.1. Mean naming latencies as a function of the type of phraseand grammatical category of the target.

the priming effect might reflect facilitation in naming the nouns due to a bias inour materials that resulted in the nouns being related more semantically than theverbs to the general meaning of the sentences. To circumvent this potentialproblem, we replicated the findings of Experiment I using nonword targets thathave suffixes primarily associated with a verb or a noun. For the "verb" non-words, we used the suffix -ed, which is a salient inflectional marker for pasttense. For the "noun" nonwords, we used the suffix -er, which is the deriva-tional morpheme for agentive. 3 An example of a "verb" and a "noun" nonwordare glatted and latter, respectively (see Table 9.2 for additional examples).Thus, the -ed and -er nonwords functionally parallel the verb and noun targetsused in Experiment 1. A syntactic priming effect would be reflected, therefore,by longer naming latencies to the -ed nonwords compared to the -er nonwordswhen they follow the sentence contexts that we used previously.

The results from this experiment are shown in Fig. 9.2. As in Experiment 1, asyntactic priming effect was found with both literal and figurative sentences.Moreover, the effect for figurative sentences was statistically larger than that forliteral sentences (49 ms vs. 17 ms, respectively). Thus, the results from thisexperiment again support the conclusion that idiomatic phrases undergo a full

3 0f course, the suffix -er also is used with comparative adjectives. However, the ambiguity in theinterpretation of the -er suffix is not a problem in the present study, because both nouns and adjectivesare syntactically appropriate continuations of our sentence contexts.

211

I

---v -


E 700C

G) 690'c0Ca)X 6800)C

E 670'coCC0) 660

Target Type

* "Verb"M "Noun"

Idiom Literal

Type of Phrase

FIG. 9.2. Mean naming latencies ,as a function of the type of phraseand the type of nonword target.

syntactic analysis. Because the targets in this experiment were meaninglessnonwords, the priming effects cannot be attributed to the semantic relations thatheld between the targets and sentence contexts.

In both Experiments 1 and 2, the syntactic priming effect for figurative primeswas larger than the effect for literal primes, although this difference was statis-tically significant only in the second experiment. Why might such an effectoccur? We assume that the representation of an idiom in memory includes infor-mation regarding both its meaning (i.e., that kick the bucket means "to die"), aswell as its syntactic structure. When one encounters the idiom in a biasingcontext, it is likely that the mental representation of the idiom will be accessedrelatively quickly (probably before one has been exposed to the entire idiom).The retrieval of the idiom will make available the idiom's syntactic structure, andthus will allow the comprehension system to project the upcoming structure ofthe remainder of the idiomatic phrase in a relatively determinate fashion. Aliteral phrase, because it does not have a preexisting mental representation, maybe less efficient at generating specific syntactic expectations. Idioms, therefore,might make available syntactically relevant information more effectively than doliteral phrases, and as a result they might be more susceptible to interferencewhen syntactic violations occur.

As noted previously, Gibbs and Gonzales (1985) presented evidence suggest-ing that syntactically frozen idioms are easier to process than are unfrozen

uuu

I

I


idioms. One possible explanation for this processing difference is that frozenidioms do not undergo a full syntactic analysis during comprehension. If thisview is correct, then one might expect that frozen idioms would be the leastsusceptible to syntactic priming effects. However, the fact that we obtained, inExperiment 2, a larger syntactic effect for idioms as compared to literal phrasessuggests that this possibility may not be correct. Even relatively unfrozen idiomstend to be less flexible syntactically than their literal counterparts, and hence onewould have expected that the syntactic effect for idioms would have been smallerthan the effect for literal phrases, regardless of the overall level of frozenness ofthe idioms.

To test directly whether frozenness affects the syntactic processing of idi-omatic strings, we obtained frozenness ratings for each of our idioms. To do this,we gave subjects either the literal or idiomatic sentence from each of our sentencepairs, along with a syntactically transformed version of the sentence. The trans-formation was either the passive, where possible, or other movement of a nounphrase (NP) in cases where passivization was not grammatical. The subject ratedhow similar in meaning the original and transformed sentences were. A 5-pointrating scale was used, with I being "very different in meaning" and 5 being"very similar in meaning." Overall, the mean rating for the idiomatic sentenceswas significantly lower than for the literal sentences (2.67 vs. 3.18, respec-tively). Thus, it is clear that the idiomatic phrases that we used are less flexiblesyntactically than are the corresponding literal phrases.

Based on the aforementioned ratings, we obtained frozenness scores for eachof our idioms by subtracting the mean rating for each idiomatic sentence from themean rating for its literal control. Thus, the larger the frozenness score, thegreater the frozenness of the idiom (i.e., a large score indicates that the ratingwas relatively low for the idiom, relative to the rating for the idiom's literalcontrol). The frozenness scores for the idioms ranged from -. 69 to 3.00. Wecomputed the correlation between the frozenness scores and the syntactic prim-ing effects obtained for the idioms in Experiments I and 2. The correlations were-. 01 and -. 08, respectively. Neither of these correlations approached signifi-cance. Thus, frozenness did not affect the magnitude of the syntactic primingeffect. Apparently, on-line syntactic expectations are not influenced by the de-gree of structural flexibility of a given idiomatic phrase. It is important to keep inmind, however, that this conclusion specifically applies to expectations of thefinal word of an idiom, presented in a strongly biasing sentence context. Perhapsfrozenness might play a role at earlier points in the idiom, and in contexts that areless constraining.

In summary, we found a syntactic priming effect for idioms with both wordand nonword targets. It appears, therefore, that the processing of idiomatic andliteral strings is largely indistinguishable in terms of the syntactic representationsthat are derived.

213


Literal Interpretation of Idiomatic Phrases

In a final experiment, we addressed the question of whether the literal meaningof an idiomatic phrase is computed, even in a context biasing the idiom's figur-ative interpretation. As sentence primes, we used those idiomatic and literalsentences from the previous experiments that required a concrete noun as acompletion. For example, in the literal context "The soccer player slipped whenhe tried to kick the . . . ," the completion must be literally "kickable," and somust be concrete. Target words in the experiment, then, were either unrelatedabstract or concrete nouns (e.g., truth versus shelf, respectively; see Table 9.1for further examples). We assumed that the semantic bias of the literal sentenceswould delay naming responses to the abstract nouns (relative to the concretenouns). We refer to such an effect as conceptual priming. Of primary interest inthis experiment, then, was whether figuratively biased sentences would show asimilar priming effect. Consider for example the sentence context "The man wasold and feeble and it was believed that he would soon kick the...." Althoughthe expected completion (bucket) is a concrete noun, the concreteness of thecompletion is essentially irrelevant to the idiom's figurative meaning per se.Thus, one would not expect to find a conceptual priming effect (i.e., slowerresponses to abstract than to concrete targets) with these sentences unless theliteral meaning of the idiom was derived along with the idiom's figurative in-terpretation. The presence of conceptual priming therefore was used in Experi-ment 3 as an indicator of ongoing literal computation.

The syntactic dominance model claims that the semantic processor obli-gatorily must analyze the literal meaning of an idiom based on structural inputfrom the syntactic processor. This model predicts, therefore, that conceptualpriming should occur with idiomatic sentences. In contrast, the syntactic-seman-tic autonomy model assumes that a full literal interpretation of an idiom is notderived, even though its syntactic structure is computed. Thus, this latter modelpredicts that conceptual priming will not arise with idiomatic sentences. Recallthat the semantic dominance model was incompatible with the results fromExperiments I and 2 (because it predicts that syntactic analysis can be terminatedonce the figurative meaning of the idiom has been retrieved). Ruling out themodel, however, requires demonstrating that the literal meaning of the idiom nolonger was being considered at the time that target word was presented. Failure tofind a conceptual priming in the present experiment therefore would serve asconvincing evidence against this model.

The same cross-modal priming procedure that was used in the previous twoexperiments was used in Experiment 3. The results of the experiment are shownin Fig. 9.3. Overall, abstract targets were named significantly more slowly thanconcrete targets. However, this conceptual priming effect interacted with the typeof sentence context. Although literal sentences showed a significant 21-ms ef-fect, the priming effect was not significant for idiomatic sentences (the effect


E 560C

co 550Ca)X 540.'atE 530CCco

4) 520-Target Type

* Abstract1 Concrete

iaiom Uterai

Type of Phrase

FIG. 9.3. Mean naming latencies as a function of the type of phraseand concreteness of the target.

with idioms was only 2 ms, and it was in a direction opposite to that found withliteral sentences). The pattern of data from Experiment 3 suggests, therefore, thatthe literal meaning of an idiomatic phrase is computed only in literally biasingcontexts. When the context biases an idiomatic interpretation of the phrase,literal analysis appears to be terminated prior to the final word of the idiom.

The results from this experiment stand in sharp contrast to the pattern ofresults found in the first two experiments. In the prior experiments, idiomaticsentences showed syntactic priming effects that were as large as (if not largerthan) literal sentences. In the present experiment, however, a conceptual primingeffect was found for literal sentences only. Taken together, these results suggestthat, by the final word of an idiomatic phrase, subjects continue to monitor thesyntactic structure of the idiom, but do not compute a corresponding literalinterpretation of the phrase. This dissociation between syntactic and semanticanalyses serves as strong evidence in support of the syntactic-semantic autonomymodel of processing. According to this model, the semantic processor terminatesits analysis of the literal meaning of an idiomatic phrase once the figurativemeaning of the idiom has been retrieved. There is, however, no feedback fromthe semantic to the syntactic processor in this model, hence the semantic pro-cessor cannot suppress directly further syntactic processing. Therefore, a fullsyntactic description of an idiom always will be computed, regardless of con-textual constraints, and regardless of the status of ongoing literal processing.These results are compatible with modular views of parsing, which stipulate that

215

II

OIUI

_ ------


the parser is insensitive to semantic and pragmatic contingencies (e.g., Frazier,Clifton, & Randall, 1983).

NEUROLOGICAL DISSOCIATIONS IN IDIOMCOMPREHENSION

In the preceding section, we presented evidence suggesting that syntactic pro-cessing occurs during the comprehension of idiomatic phrases. We also demon-strated that, in a figuratively biasing context, the literal interpretation of anidiomatic phrase is not derived. This dissociation between syntactic and semanticlevels of analysis is important, and will need to be accounted for by any completemodel of idiom comprehension. Intriguingly, such dissociations have relevantparallels in the neurolinguistic literature. In this section of the chapter, we con-sider neurolinguistic studies of idiom comprehension, and attempt to relate thatdata to the priming data we presented in the last section. Our discussion focusesspecifically on the nature of hemispheric asymmetries in the processing of literaland idiomatic language.

In the past decade we have attained a notable degree of neurolinguistic sophis-tication beyond the stereotype that (right-handed) humans are strongly lateralizedfor language in the left hemisphere and lateralized for visual/spatial processing inthe right hemisphere. Although there is a consensus that language production isstrongly left-lateralized, it is now less controversial to acknowledge that the righthemisphere possesses certain language comprehension abilities. There is an in-creasing amount of evidence that suggests that both the right and left hemi-spheres contribute to the comprehension of semantic relationships (Burgess &Simpson, 1988b; Chiarello, 1988a; Van Lancker, 1985). For example, bothhemispheres show evidence of a basic semantic priming effect with words thatare associatively and semantically related (e.g., doctor-nurse; Chiarello, 1985).In fact, the right hemisphere actually shows an advantage over the left hemi-sphere in priming for words that have weak or indirect semantic relations (e.g.,dull-moody, lawyer-nurse; Burgess & Rosen, 1988; Chiarello, Burgess, Rich-ards, & Pollock, 1990). Similarly, although dominant and subordinate meaningsof ambiguous words show priming in either hemisphere, the subordinate mean-ings stay active longer in the right hemisphere (Burgess & Simpson, 1988a).Thus, it has become recognized increasingly that both hemispheres have contri-butions to make to the lexical/semantic processing required by the languagecomprehension system.

A more specialized account of hemispheric capability emerges, however,when we consider the language subsystems that underlie the processing of syn-tactic structure one the one hand, and figurative processing on the other. Thedomain of syntactic processing appears to be more clearly a left-hemispherefunction. For example, patients with left-hemisphere damage seem unable to


distinguish between such sentences as "The boy kisses the girl" and "The girlkisses the boy," a distinction that requires the coordination of the meaning ofindividual lexical units with the serial syntactic structure of the sentence(Gazzaniga & Hillyard, 1971; Heeschen & Jurgens, 1977). Syntactic com-prehension difficulties with left-hemisphere-damaged patients are most pro-nounced when the context does not provide any semantic or discourse-levelconstraints for syntactic interpretation (Caplan & Hildebrandt, 1988; Caramazza& Berndt, 1978; Caramazza & Zurif, 1976).

Conversely, figurative language seems most impaired with patients who expe-rience right-hemisphere brain damage. For example, these patients have diffi-culty appreciating the connotative relationships between words (Brownell, Pot-ter, Michelow, & Gardner, 1984), making inferences (Beeman, 1990),understanding indirect speech acts (Foldi, 1987; Hirst, Le Doux, & Stein, 1984),appreciating humor (Bihrle, Brownell, Powelson, & Gardner, 1986; Wapner,Hamby, & Gardner, 1981), comprehending metaphors (Brownell, 1988), as wellas understanding idiomatic phrases (Myers & Linebaugh, 1981).

An experiment by Van Lancker and Kempler (1987) directly explored theneural underpinnings of figurative and literal comprehension. These researchershad left-hemisphere-damaged (LHD) and right-hemisphere-damaged (RHD) pa-tients select the appropriate line drawing (out of a set of four) that correspondedto either a single concrete noun (bird), a sentence containing a figurative phrase("He's turning over a new leaf"), or a novel literal sentence ("He's sitting deepin the bubbles"). For the idioms, one drawing was related to the literal interpreta-tion of the idiom, whereas another was consistent with the idiom's figurativeinterpretation. The other two drawings depicted a scene that was either oppositein meaning to the idiom or was irrelevant. The drawings used with the novelliteral sentences were somewhat different. All three of the distractors for thesesentences involved some change in grammatical or thematic roles of the nouns(e.g., agent-patient reversal) or misassignment of adjective modification (e.g.,the sentence "When the happy girl pushes, the angry boy swings" had a drawingof an unhappy girl pushing a boy). Correct performance on the literal sentences,therefore, required careful attention to the syntactic details of the sentences. Acritical finding from the experiment was that there was an interaction betweenpatient type and stimulus set. Specifically, LHD patients performed worse thanRHD patients on the novel literal sentences (those that required detailed syntacticanalysis), but outperformed the RHD patients on the figurative sentences. Appar-ently, left-hemisphere damage interferes significantly with syntactic analysis, butis less disruptive of figurative processing. Based on these results, Van Lanckerand Kempler argued that syntactic analysis is not a necessary component ofidiomatic interpretation. They suggested that the results are supportive of an"idiom-as-a-word" hypothesis or what we earlier referred to as the lexical repre-sentation hypothesis (Swinney & Cutler, 1979). According to this view, anidiom's meaning is directly (and holistically) retrieved from memory (presum-

217


ably from the right hemisphere) without any consideration of the idiom's internalstructure.

The claim that idiom comprehension does not require syntactic analysis hasimportant implications for the data that we presented in Experiments 1 and 2.Recall that we demonstrated that the magnitude of syntactic priming that occurswith idiomatic strings is equal to (or even surpasses) that found with literalphrases. Van Lancker and Kempler's (1987) conclusions suggest, therefore, thatthe priming effects we observed may not be directly relevant to the retrieval offigurative meaning per se. Instead, the priming effects simply may reflect obliga-tory parsing of the idiomatic phrases by the left hemisphere. If this is correct, itpotentially explains why we observed independence between syntactic and se-mantic analyses. Perhaps analysis of the figurative meaning of an idiom occurs inthe right hemisphere, neurally dissociated from left-hemisphere parsing.

Although the idiom-as-a-word hypothesis is intriguing, there are, neverthe-less, a number of empirical problems it faces. The first comes from Van Lanckerand Kempler's (1987) own data. Recall that their experiment used a single-wordcondition in addition to the idiom and literal sentence conditions. Performancedid not differ reliably between LHD and RHD groups in the single word condi-tion and, critically, their performance in this condition was very close to that ofintact control subjects. These data indicate, therefore, that the patients werecapable of retrieving lexical items from memory. If an idiom truly is representedas a simple lexical item (as the idiom-as-a-word hypothesis suggests) then onewould expect that the patients would be highly competent in comprehendingthese items as well. In particular, one would expect the LHD patients to performin the idiomatic condition at roughly the same level as they did with singlewords, and at approximately the same level as normal subjects. This pattern didnot occur, however. Although LHD patients outperformed their RHD counter-parts, their performance was still significantly worse than the normal controls,and was worse than their own performance in the single-word condition. Clearly,then, idiomatic interpretation involves processing that is in some way morecomplex than a simple lexical representation view would propose. Perhaps thedecrement in performance in the idiomatic condition by LHD patients resultedfrom their impoverished ability to analyze the idiomatic phrases structurally. Ifthis claim (which is admittedly speculative) turns out to be correct, it wouldsuggest, in contrast to the idiom-as-a-word hypothesis, that syntactic analysis isan important component of idiomatic processing.

A second important challenge to the idiom-as-a-word hypothesis comes fromthe results that we presented in Experiment 2. In that experiment, the primingeffect for idiomatic phrases was significantly larger than that for literal phrases.We suggested that such a pattern might be expected to occur if the stored repre-sentation of an idiom contains information regarding the phrase's syntactic struc-ture. Such information might be made available during the processing of thephrase, and then could be used by the parser to generate specific syntactic


expectations. If the idiom-as-a-word hypothesis is correct, it become a mysteryas to why the representation of an idiom would guide syntactic processing in thisway. Presumably, syntactic information would be stored with an idiom only ifthat information were relevant to the idiom's retrieval. Thus, we suggest that ourdata support a view of idiom processing in which structural analysis contributesto successful comprehension. Further, given that we found no effect of syntacticfrozenness on the magnitude of syntactic priming, it appears that syntactic analy-sis contributes to the comprehension of both frozen and unfrozen idioms.

Given this view, how might we account for Van Lancker and Kempler's(1987) finding that LHD patients (who presumably have impaired syntacticcapabilities) performed better on idiomatic sentences than did RHD patients(who have intact parsers)? There may have been two factors that conspired toproduce this result. First, it may be that the RHD patients had difficulty with theidiomatic strings due to particular processing biases of the left hemisphere.Specifically, the left hemisphere has been characterized as making rapid andinflexible commitments when interpreting lexical items (Brownell, Potter,Bihrle, & Gardner, 1986; Burgess & Simpson, 1988b; Chiarello, 1988b). Thus,RHD patients may have interpreted the initial words of the idiom literally, and, asa result, may have had trouble appreciating the figurative status of the remainderof the phrase. According to this view, then, the left hemisphere has difficultywith idiomatic phrases because it is "garden-pathed" away from a figurativeinterpretation during processing. The right hemisphere does not have this kind oftrouble with idiomatic phrases because it tends not to make rigid interpretivecommitments (Brownell et al., 1986; Burgess & Simpson, 1988b; Chiarello,1988b). Note that this account of hemispheric asymmetries is qualitatively differ-ent from that given by the idiom-as-a-word hypothesis. According to that hy-pothesis, RHD patients have difficulty with idiomatic phrases because idiomaticentries in the lexicon occur only in the right hemisphere. The obvious problemwith this argument is that there is no a priori account of why the left hemisphereis unable to represent idiomatic words. Our account, in contrast, focuses noton representational differences between the hemispheres, but instead on well-documented processing differences. As a result, our hypothesis fits more directlyinto extant neurolinguistic models of comprehension.

Although the aforementioned hypothesis accounts for the poor performanceby RHD patients on the idiomatic phrases, we still must explain why the LHDpatients (whose parsers apparently were impaired) were able to comprehend theidioms at a relatively high level (although at a level still below that of normalsubjects). Research by Caramazza and Zurif (1976) may provide an answer tothis question. These investigators demonstrated that, although LHD patientshave difficulty interpreting sentences such as "The tiger is chasing a fat lion,"they performed much better on sentences like "The bicycle that the boy isholding is broken." The critical difference between these sentences appears to bethat, in the first, the subject and object noun phrases are reversible (in the sense


that either pragmatically could be the agent of the verb), whereas in the secondsentence the nouns are nonreversible (because bicycle is an inanimate object andthus can not serve as the agent of the sentence). These data suggest that the righthemisphere is adept at utilizing nonsyntactic cues to derive the structural rela-tions among the components of a sentence. Perhaps such analyses were used byLHD patients in the Van Lancker and Kempler (1987) experiment to determinethe structure of idiomatic phrases as well. Consider, for example, how the righthemisphere might analyze a sentence like "He kicked the bucket." Because thesubject and object noun phrases of this sentence are literally nonreversible (i.e.,bucket cannot be the agent of kicked because it is an inanimate noun), the righthemisphere presumably could determine the underlying literal structure of thissentence without syntactic mediation. The retrieval of this structure subsequentlymight allow access to the idiom's representation in memory (which itself con-tains the information that bucket is the underlying object of the phrase). Accord-ing to this view, then, the right hemisphere sometimes might be able "bootstrap"its way to a figurative interpretation by first deriving a thematic representation ofan idiomatic phrase. This bootstrapping hypothesis predicts that LHD patientsmight be particularly sensitive to the literal relations among the components of anidiomatic phrase. This prediction could be tested empirically using the concep-tual priming paradigm that we outlined in Experiment 3. We would expect thatLHD patients might show a robust conceptual priming effect for idiomaticphrases. Such a finding would contrast with the data from our (intact) subjects,who did not display conceptual priming when idiomatic phrases were presented.The bootstrapping hypothesis also predicts that LHD patients might be easilyconfused by idioms whose nouns are literally reversible.

In our experiments, we demonstrated that the syntactic structure of an idiom isobligatorily computed, even though its literal meaning is not always derived.How might we accommodate these findings within a neurolinguistic model ofcomprehension? We suggest the following processing scenario. When presentedwith linguistic input, both hemispheres attempt, in parallel, to map individualword meanings onto an evolving discourse representation. The kinds of informa-tion that they have available to them lo perform this task differs, however. Inparticular, the left hemisphere can make use of syntactic constraints whereas theright hemisphere must build interpretations on the basis of nonstructural cuesalone. We assume that, in general, the left hemisphere is more efficient atintegrating information into the discourse model, and thus it will tend to controlinterpretation. However, because the left hemisphere tends to make rapid, andinflexible, commitments to particular interpretations (at the expense of alter-native, more esoteric, possibilities), it sometimes will fail to appreciate somepotential aspects of meaning. As noted. previously, idiomatic meanings may beparticularly problematic for the left hemisphere in this regard, because the lefthemisphere may be biased to interpret the initial words of an idiom literally, andmay be unable to recover from these initial commitments. Thus, in general, the


left hemisphere may be biased to compute a literal interpretation of an idiomaticstring, possibly even in figuratively biasing contexts. The fact that we found, inExperiment 3, that the literal meaning of an idiom was not fully computed in abiasing context implies, therefore, that the right hemisphere sometimes can usurpleft-hemisphere interpretive processing, and can integrate idiomatic interpreta-tions into the discourse representation.

What processing event might trigger the right hemisphere to contribute todiscourse processing in this way? Perhaps as the left hemisphere is evaluating theliteral meaning of an idiomatic string, it detects that the meaning is anomalouswith the preceding (figuratively biased) context. The detection of anomaly, then,might cause an attentional shift from the left to the right hemisphere, as thesystem searches for an acceptable interpretation of the phrase. The shift inattention could make the figurative meaning available, and thus allow it to beintegrated into the discourse representation. Successful integration, then, couldcreate "closure" at the discourse level, blocking further attempts to map themeanings of the individual words of the idiom into the discourse model. As aresult, literal processing in the left hemisphere might be aborted (accounting forthe lack of a conceptual priming effect for idioms in our third experiment).Closure at the discourse level, however, might not disable lower level systems inthe left hemisphere. Thus, parsing of the idiomatic phrase could continue, evenafter figurative integration is complete.

It is important to note that the views just discussed does not necessarily implythat retrieval of the figurative meaning of an idiom will occur only after the entirephrase has been evaluated literally by the left hemisphere. If detection of ananomaly occurs during processing of the initial words of the phrase, then re-trieval and integration of the figurative meaning of the idiom could take placebefore the final words of the idiom are processed fully. Thus, the anomaly viewis, in fact, compatible with data showing that the retrieval of idiomatic meaningoccurs as fast as (or perhaps faster than) the derivation of a phrase's literalmeaning (e.g., Ortony et al., 1978). It is important to note, however, that theanomaly view predicts that a delay in figurative processing could occur if theinitial parts of the idiomatic phrase can be integrated literally into the precedingcontext (i.e., if there is no cue that the literal interpretation is incorrect). Cacciariand Tabossi (1988) recently presented evidence consistent with this prediction.

Although anomaly detection might contribute to idiomatic interpretation, it isprobably not a necessary component. Consider, for example, the sentence con-text "The man was old and feeble and it was believed that he would soonkick...." Intuitively, one assigns a figurative interpretation to kicked in thiscontext, even though the literal meaning of the word is not, in any obvious way,anomalous with the preceding context. To account for this phenomenon, wesuggest that the sentence context primes the idiom's representation in memory (inmuch the same way that a sentence context might prime a semantically relatedword), and thereby increases the speed by which the right hemisphere can re-

221


trieve the idiomatic representation. Early retrieval might allow the right hemi-sphere to integrate a figurative meaning into the discourse representation beforethe left hemisphere has had time to commit to a literal interpretation. This viewsuggests, therefore, that the two hemispheres compete with one another for theprivilege of integrating their particular interpretations into a discourse structure.The competition, then, is of a "horse-race" variety, with the winner being thehemisphere that is the first to develop a contextually appropriate interpretation ofthe input. Under most circumstances, the left hemisphere wins this race easily.However, with idiomatic phrases presented in a biasing context, the right hemi-sphere sometimes can prevail as well.

Interestingly, the horse race model predicts that the detection of a literalanomaly can influence idiom processing as well. That is, according to the model,a figurative interpretation of an idiomatic phrase will be derived any time theright hemisphere can retrieve that interpretation before the left hemisphere canintegrate words literally. Logically, such a right-hemisphere advantage could beobtained by either speeding up processing in the right hemisphere (e.g., primingthe idiomatic meaning) or slowing processing in the left hemisphere. The detec-tion of anomaly, in fact, may have this latter influence. That is, words that areliterally anomalous with the preceding context will force a delay in integration inthe left hemisphere, thereby providing time for the right hemisphere to retrieve apreferable (i.e., figurative) interpretation. The horse race model suggests, there-fore, that context can have two mutually reinforcing effects on idiom comprehen-sion. Specifically, context can both provide support for a figurative interpretationand, simultaneously, hinder literal integration (see McDonald & Carpenter,1981, for a similar proposal).

SUMMARY

In this chapter, we have focused on interdependencies that exist between differentlinguistic processing systems during the course of understanding idiomaticspeech. In the first section of the article, we explored specifically the processingrelationship that exists between semantic and syntactic analyses, whereas in thefinal section we explored how these levels of processing might be coordinatedacross the two cerebral hemispheres. We demonstrated, in a set of three experi-ments, that the full syntactic structure of an idiomatic phrase is computed in afiguratively biasing context, even when the phrase's literal meaning no longer isbeing evaluated. These results serve as clear evidence in support of a processingdistinction between syntax and semantics, and highlight the obligatory characterof structural analysis per se. We presented a neural processing model, then, toaccount for these results. According to this model, the integration of a figurativemeaning in the right hemisphere into a discourse representation blocks furthersemantic evaluation in the left hemisphere. Figurative integration, however, does


not (or cannot) interfere with left-hemisphere parsing, perhaps because the righthemisphere is ill-equipped to communicate with the parser in such a fashion.Further research clearly is needed to elaborate on the role of semantic andsyntactic analyses in the retrieval of figurative meaning. We believe that the-oretical advancement will come most efficaciously, however, if considerations ofhemispheric processing serve as a springboard for this research.

REFERENCES

Altman, G. T., & Steedman, M. (1988). Interaction with context during human sentence process-ing. Cognition, 30, 191-238.

Beeman, M. (1990). Coherence inferencing and structure building in the cerebral hemispheres.Unpublished doctoral dissertation, University of Oregon, Eugene.

Birlfe, A. M., Brownell, H. H., Powelson, J. A., & Gardner, H. (1986). Comprehension ofhumorous and non-humorous materials by left and right brain-damaged patients. Brain andCognition, 5, 399-411.

Brownell, H. H. (1988). Appreciation of metaphoric and connotative word meaning by brain-damaged patients. In C. Chiarello (Ed.), Right hemisphere contributions to lexical semantics (pp.19-31). New York: Springer-Verlag.

Brownell, H. H., Potter, H. H., Bihrle, A. M., & Gardner, H. (1986). Inference deficits in rightbrain-damaged patients. Brain and Language, 29, 310-321.

Brownell, H. H., Potter, H. H., Michelow, D., & Gardner, H. (1984). Sensitivity to lexical denota-tion and connotation in brain-damaged patients: A double dissociation? Brain and Language, 22,253-265.

Burgess, C., & Rosen, A. (1988, April). Priming of emotional words in the cerebral hemispheres.Paper presented at the Fifth Annual Linguistic Studies Conference on Language and Communica-tion, Syracuse, NY.

Burgess, C., & Simpson, G. B. (1988a). Cerebral hemispheric mechanisms in the retrieval ofambiguous word meanings. Brain and Language, 33, 86-104.

Burgess, C., & Simpson, G. (1988b). Neuropsychology of lexical ambiguity resolution: The contri-bution of divided visual field studies. In S. L. Small, G. W. Cottrell, & M. K. Tanenhaus (Eds.),Lexical ambiguity resolution in the comprehension of human language (pp. 411-430). Los Altos,CA: Morgan Kaufmann.


Caplan, D., & Hildebrandt, N. (1988). Disorders of syntactic comprehension. Cambridge, MA:MIT Press.

Caramazza, A., & Berndt, R. S. (1978). Semantic and syntactic processes in aphasia: A review ofthe literature. Psychological Bulletin, 85, 898-918.

Caramazza, A., & Zurif, E. B. (1976). Dissociation of algorithmic and heuristic processes inlanguage comprehension: Evidence from aphasia. Brain and Language, 3, 572-582.

Chiarello, C. (1985). Hemisphere dynamics in lexical access: Automatic and controlled priming.Brain and Language, 26, 146-172.

Chiarello, C. (Ed.). (1988a). Right hemisphere contributions to lexical semantics. New York:Springer-Verlag.

Chiarello, C. (1988b). Semantic priming in the intact brain: Separates roles for the right and lefthemispheres? In C. Chiarello (Ed.), Right hemisphere contributions to lexical semantics (pp. 59-69). New York: Springer-Verlag.


Chiarello, C., Burgess, C., Richards, L., & Polloek, A. (1990). Semantic and associative primingin the cerebral hemispheres: Some words do some words don't, . . . sometimes, some places.Brain and Language, 38, 75-104.

Crain, S., & Steedman, M. (1985). On not being led up the garden path: The use of context by thepsychological parser. In D. Dowty, L. Kartunnen, & A. M. Zwickey (Eds.), Natural languageparsing (pp. 320-358). Cambridge, England: Cambridge University Press.

Estill, R. B., & Kemper, S. (1982). Interpreting idioms. Journal of Psycholinguistic Research, 6,559-568.

Ferreira, F., & Clifton, C. (1986). The independence of syntactic processing. Journal of Memoryand Language, 25, 348-368.

Foldi, N. S. (1987). Appreciation of pragmatic interpretations of indirect commands: Comparisonof right and left hemisphere brain-damaged patients. Brain and Language, 31, 88-108.

Fraser, B. (1970). Idioms within a transformational grammar. Foundations of Language, 6, 22-42.Frazier, L., Clifton, C., & Randall, J. (1983). Filling gaps: Decision principles and structure in

sentence comprehension. Cognition, 13, 18W-222.Gazzaniga, M. S., & Hillyard, S. A. (1971). Language and speech capacity of the right hemi-

sphere. Neuropsychologia, 9, 273-280.Gibbs, R. (1986). Skating on thin ice: Literal meaning and understanding idioms in conversation.

Discourse Processes, 9, 17-30.Gibbs, R., & Gonzalez, G. P. (1985). Syntactic f]-ozenness in processing and remembering idioms.

Cognition, 20, 243-259.Goodman, G. O., McClelland, J. L., & Gibbs, R. W. (1981). The role of syntactic context in word

recognition. Memory & Cognition, 9, 580--586.Heeschen, C., & Jurgens, R. (1977). Pragmatic-semantic and syntactic factors influencing ear

differences in dichotic listening. Cortex, 13, 74-84.Hirst, W., Le Doux, J., & Stein, S. (1984). Constraints on the processing of indirect speech acts:

Evidence from aphasiology. Brain and Language, 23, 26-33.Marslen-Wilson, W. D. (1985). Speech shadowing and speech communication. Speech Communi-

cation, 4, 55-73.Marslen-Wilson, W. D., & Tyler, L. K. (1980). The temporal structure of spoken language under-

standing. Cognition, 8, 1-71.McDonald, J. L., & Carpenter, P. A. (1981). Simultaneous translation: Idiom interpretation and

parsing heuristics. Journal of Verbal Learning and Verbal Behavior, 20, 231-247.Myers, P. S.. & Linebaugh, C. W. (1981). Comprehension of idiomatic expressions by right-

hemisphere-damaged adults. In R. H. Brookshire (Ed.), Clinical aphasiology: Conference pro-ceedings, 1981. Minneapolis: BRK.

Ortony, A., Shallert, D. L., Reynolds, R. E., &: Antos, S. (1978). Interpreting metaphors andidioms: Some effects of context on comprehension. Journal of Verbal Learning and VerbalBehavior, 17, 465-477.

Peterson, R. R., Burgess, C., Dell, G. S., & Elberhard, K. (1989, March). Dissociation of syntacticand semantic analyses during idiom processing. Paper presented at the second annual CUNYConference on Human Sentence Processing, New York.

Rayner, K., Carlson, M., & Frazier, L. (1983). The interaction of syntax and semantics duringsentence processing: Eye movements in the analysis of semantically biased sentences. Journal ofVerbal Learning and Verbal Behavior, 22, 3.5 -374.

Reagan, R. T. (1987). The syntax of English idioms: Can the dog be put on? Journal of Psycho-linguistic Research, 16, 417-441.


Taraban, R., & McClelland, J. L. (1988). Constituent attachment and thematic role assignment insentence processing: Influences of content-based expectations. Journal of Memory and Lan-guage, 27, 597-632.


Van Lancker, D. R. (1985). Nonpropositional speech: Neurolinguistic studies. In A. W. Ellis (Ed.),Progress in the psychology of language (Vol. 3, pp. 49-118). Hillsdale, NJ: Lawrence ErlbaumAssociates.

Van Lancker, D. R., & Kempler, D. (1987). Comprehension of familiar phrases by left- but not byright-hemisphere damaged patients. Brain and Language, 32,265-277.

Wapner, W., Hamby, S., & Gardner, H. (1981). The role of the right hemisphere in the apprehen-sion of complex linguistic materials. Brain and Language, 14, 15-33.

West, R. F., & Stanovich, K. E. (1986). Robust effects of syntactic structure on naming. Memory &Cognition, 14, 104-112.

Wright, B., & Garrett, M. (1984). Lexical decision in sentences: Effects of syntactic structure.Memory & Cognition, 12, 31-45.

I I I MEANING AND STRUCTURE

227

O ^Building Castles in the Air.1 XSome Computational andTheoretical Issues inIdiom Comprehension'

Oliviero StockJon SlackIRST. Istituto per la Ricerca Scientifica e Techologica, Italy

Andrew OrtonyInstitute for the Learning Sciences, Northwestern University

A successful natural language understander, be it a human or a machine, fre-quently is faced with the problem of making sense of linguistic expressions thatdo not necessarily mean what they say. There are many different kinds of ex-pressions that have this property, some of which have received more attention inthe research literature than others. There has been, for example, a great deal ofwork on the processes involved in the comprehension of indirect speech acts(e.g., Clark & Lucy, 1975; Gibbs, 1983) and there is an even larger literature onthe comprehension of metaphorical uses of language (e.g., Ortony, in press).However, until recently, the comprehension of idioms has been studied lessextensively, partly perhaps because there is a rather appealing and intuitive viewof the nature of idioms that has the effect of depriving the question of how theyare understood of any really interesting or problematic properties. We refer tothis as the "idioms-are-big-words" view.

As recently as a decade ago research on idiom comprehension (e.g., Ortony,Schallert, Reynolds, & Antos, 1978; Swinney & Cutler, 1979) accepted quiteuncritically the idioms-are-big-words view. Swinney and Cutler proposed a pro-cess model in which idioms are accessed from the lexicon in parallel with anattempt at literal interpretation. According to the model, familiar idioms arelikely to be processed very quickly because as "long words" they do not requireany deep linguistic analysis-they simply are recognized. From this view thecomprehension of an idiom may be accompanied by some partial processing of

'This work has been carried on as part of MAIA, an experimental platform of the integrated Alproject under development at IRST.

229

230 STOCK, ORTONY, SLACK

the literal meaning, but the literal meaning plays no role in the determination ofthe idiomatic meaning.

Consider the following examples:

1. The car John bought was a pig in a poke.2. The ice was unexpectedly broken when John entered the room.

Example I is no problem for a language understander. There is an Englishidiom (or part of one) pig in a poke, which never occurs in any other form. So,for example, you may have bought a pig in a poke this morning but you cannotsay that you bought a pig this morning in a poke (although you might well havebought a pig this morning in an auction ). Similarly, the pig in a poke that Johnbought he presumably bought unwittingly, but he could not be said to havebought a pig unwittingly in a poke. The phrase pig in a poke simply cannot bemanipulated in any way whatsoever, either lexically or syntactically. This meansthat it can be entered safely in the lexicon as a lexical item-a noun with itsassociated meaning.

One might take the position that the expression pig in a poke is in fact onlypart of an idiom-that the "real" idiom is the verb phrase "buy a pig in a poke."If one takes this view, then one has to deal with the possibility that not only canone "buy a pig in a poke," but that one might also get away with "selling a pig ina poke" as well as "getting a pig in a poke." In this case one would have topropose a small group of pig-in-a-poke idioms, some of which (certainly the firsttwo cases) allow syntactic changes. For example, one can say "The pig in a pokethat we're talking about I bought yesterday." Because there is no objective wayto select between the two approaches, it seems preferable to take the noun phraseas the idiom and to specify in its representation the verb forms with which it canappear.

Wood (1986) defined idioms as fixed expressions that lie at the end of twocontinua: They are totally noncompositional and totally nonproductive. From thepractical perspective of natural language understanding, such a definition is oflimited utility. This is because if an expression is totally nonproductive (bothlexically and syntactically), it is trivially easy for a natural language understanderto process. Expressions such as by and forge and hither and thither really can behandled using the idioms-are-big-words approach by simply putting them in thelexicon along with pig in a poke. The main problem for a natural languageunderstander is to recognize that it is dealing with an idiom, and this problem istrivialized if the expression occurs always and only in a unique form: Simplepattern matching will solve it. If, on the other hand, the expression has somedegree of flexibility, a simple pattern-matching mechanism will not suffice. Froma practical perspective it is no solution to define away the problem by relabelingthe computationally challenging cases so that they are no longer called idioms.Rather, we need to specify (preferably psychologically plausible) mechanisms

10. SOME COMPUTATIONAL AND THEORETICAL ISSUES

for dealing with nonliteral uses of language-especially idioms and idiomlikeexpressions. Given this as the goal, the idioms-are-big-words approach is toolimited in scope. There are just too many expressions that are not amenable tosuch an account for it to represent anything like a general solution to the idiomcomprehension problem.

TWO THEORETICAL ISSUES

The approach one takes to the question of how idioms are understood depends tosome extent on the solutions one adopts to some basic theoretical issues having todo with idioms. One of these concerns the question of whether some (or evenmost) idioms are based on underlying conceptual metaphors as, for example,Lakoff (in press), Nayak & Gibbs (1990), and Gibbs and O'Brien (1990) pro-posed. The other, related issue has to do with the question of why idioms havethe linguistic properties that they have.

The question of the nature of the relationship between literal and idiomaticmeaning is related to the general issue of compositionality in idioms. The non-compositionality criterion, which is used so frequently to characterize idioms,reflects the widespread belief that the meanings of individual constituents do notcontribute their literal meanings to the meaning of an idiom as a whole (e.g., themeanings of kick and bucket play no role in the idiomatic meaning of the idiom tokick the bucket). There are reasons, however, to suppose that this is not alwaystrue. The extreme version of the rejection of this view can be found in, forexample, the work of Glucksberg (e.g., 1990), who argued that the literal mean-ings of the constituents must make a contribution because they can be exploitedin creative modifications of the conventional idiom. A more conservative reser-vation is based on the fact that there is a whole class of idioms for whichnoncompositionality is false, a class that Wood (1986) referred to as "meta-phorical idioms." We have in mind here expressions such as "to cross [that]bridge when X comes to it." In expressions of this kind, the meaning in principlecould be derived from the constituents by whatever process is employed or-dinarily in understanding metaphors. In this particular case, for example, onereadily could produce a more abstract representation of its meaning-somethinglike "to engage in [that] activity when circumstances demand," and from thatone could generalize further to some abstract notion such as postponement untilnecessary. Cacciari and Glucksberg (1990) referred to idioms of this kind asanalyzable and transparent.

In general it seems plausible to suppose that for nonmetaphorical idioms-thekind that Cacciari and Glucksberg (1990) called analyzable and opaque-theapparent dissociation between the literal and idiomatic meaning is simply due tothe fact that the connection is buried in the history of the language and theculture. Thus, for example, the relation between the literal and idiomatic mean-

231


ing of spill the beans is still not arbitrary, at least not from a historical perspec-tive, but it is generally unknown and inaccessible to native speakers of English inmuch the same way as is the connection between the original and modemmeanings of the word sinister. Gibbs (Nayak & Gibbs, 1990) made much of theapparent dissociation, arguing that in fact very often such idioms are grounded indeep conceptual metaphors that motivate them and that explain why they meanwhat they do. We suspect, however, that Gibbs underestimated the importance ofthe historical underpinnings of idioms by investing in an account that imposes onmany idioms a rational/cognitive basis that they do not really possess.

Gibbs' view appears to be that a particular idiom is contextually appropriatebecause different idioms refer to different (parts of) underlying conceptual do-mains. For example, Nayak and Gibbs (1990) wrote:

Idioms are not discretely represented in the mental lexicon in such a way that thefigurative meaning of each phrase is arbitrarily linked to each idiomatic wordstring. Instead idioms whose meanings are generated from the same conceptualmetaphor may be linked together in a temporal sequence as part of a semantic fieldfor each type of conceptual referent (anger fear joy etc.). ... [This] systematicmapping is due to people's understanding of certain aspects of emotion concepts interms of familiar concrete and well-stiuctured concepts such as heat and ani-mals. . . . These conceptual metaphors . . . are used to understand and perhapscreate our conceptual knowledge of these abstract and less well-structured con-cepts. (p. 328)

We are skeptical about the role of such "conceptual metaphors" in understandingand creating (in this case emotion) concepts. If we wanted a psychologicallyplausible mechanism for understanding such concepts (and hence presumably forunderstanding language that embodies them), we would need on this accountsome sort of shared meaning representation for the idioms and their associated(metaphorically grounded) concepts. We think, however, that there are moreparsimonious ways of building a natural language understander and that the kindof data that led Gibbs and his colleagues to their conclusions are compatible withother correspondingly parsimonious explanations. At the same time, the data aresuggestive of other important facts.

The kind of findings that lead Gibbs (Gibbs & O'Brien, 1990; Nayak &Gibbs, 1990) to his view that (many) idioms have a metaphorical basis are thatthe choice among semantically equivalent idioms is context sensitive, in thatsubjects prefer idioms that are compatible with a primed underlying metaphor,and that different members of idiom families can be shown to relate to differentparts of a complex concept. In general, findings of the first kind raise thepossibility that subjects prefer consistency. If the context describes somebody asmetaphorically "growling" in anger, then an idiom such as bite X's head off ispreferred to one such as blow one's top. However, this preference might besimply the result of surface associative connections-people no doubt also


would judge the word bite to be closer than the word blow to a prime like growl,even in a situation where no idioms and metaphors were involved at all. Thesecond kind of finding shows that the related concept is complex but not that it isnecessarily metaphorical in nature. So, for example, Nayak and Gibbs (1990,Experiment 2) found that some idioms from the same family (e.g., idioms havingto do with anger) refer to different temporal parts of a typical anger sequence(e.g., something gets on one's nerves before one blows one's top, and later onemay cool down). They take this as evidence against the traditional view that eachof these idioms needs a "separately stipulated meaning" (which they propose isthe same in each case, i.e., get angry; p. 320). We are not convinced by this lineof reasoning. One easily could acknowledge that each of these idioms refers tosomething slightly different and one equally could admit readily that (most of)the difference has to do with the stage of a typical anger episode without drawingany counter intuitive or dramatic conclusions. We merely need conclude that toget on one's nerves means (something like) "to irritate one" (which is normallytemporally prior to getting angry, or at least expressing anger), that to blow one'stop means "to express strong anger," and that to cool down means "to (start the)return to normal after an expression of strong anger." All that follows from this isthat if anybody oversimplifies by asserting that all three idioms mean "to getangry" they are wrong, but most native speakers know they have slightly differ-ent meanings anyway.

On the positive side, however, we think there is something important aboutthe subtle variations between the meanings of different idioms in the samedomain. For example, it commonly is said that to kick the bucket means "to die."But we think this is an oversimplification. One could not describe a convictedmurderer who just had been given a lethal injection by a state executioner ashaving just kicked the bucket. The idiom seems to require death by (more or less)natural causes, and relatively suddenly too. But this means that the idiom carvesout a certain piece of our knowledge about dying. It is as though it imposedconstraints on the values of some of the parameters (e.g., method, cause, etc.)associated with dying. Such an account at least would explain why we have suchidioms-they serve to specialize an existing concept in a way not specialized byan existing lexical item. (This is probably not always true, but it is probably oftenor even usually true).

Although semantic and discourse productivity are interesting phenomena wedo not think that it is incumbent on a theory of idiom comprehension to accountfor them. Both Glucksberg (e.g., 1990) and Gibbs (1990) have proposed that theliteral meanings of idioms do play some role in their comprehension.Glucksberg, in criticizing the process model of Swinney and Cutler (1979),argued that idioms do not behave like long words. For example, many idioms cansurvive lexical substitution. Glucksberg illustrated this by arguing that one can"crack the ice" rather than break it. However, he failed to notice two things thatare relevant here. First is the fact that such lexical productivity, although occa-

233


sionally possible, is certainly not the rule (one cannot "drop the beans" or"release the cat from the bag"). Second, even when it is possible, the range ofsubstitutions is very restricted. For example, even if we grant the possibility of"cracking the ice," one cannot communicate the idiomatic meaning of break theice using "fracture the ice" or "break the frozen water." Thus, if one werewedded to the idioms-are-big-words view, Glucksberg's objections would berelatively easy to deal with. One could argue first that there are two linguisticallyclose realizations of some idioms (break/crack the ice or cooklroast X's goose),and second that the examples he offered exploit a metalinguistic strategy, namely,the use of a pun, the whole point of which is to say something ambiguous,where, in this instance one reading is somehow (e.g., syntactically, semantically,or pragmatically) defective. Our view of this is that such semantic and discourseproductivity phenomena in fact exploit their own bizarreness-a speaker knowsfull well that there is something wrong with talking about "cracking the ice" andthis very fact motivates his or her creative use of such an expression. At the sametime we certainly agree that such creative linguistic acts depend critically on theavailability of the literal meanings of the constituents of the idiom in order forthem to be so exploited. The model we propose allows for the simultaneousavailability of the literal meaning (at least up to a point), although it does notclaim that the literal meaning plays an (active) role in the processing of theidiomatic meaning. Being available is not the same as making a contribution.

SYNTACTIC FLEXIBILITY

Having argued against the proposal that the semantic and discourse productivityof idiomatic expressions show that the literal meanings of their constituents playa role in their comprehension, we now want to argue that the degree of syntacticflexibility exhibited by idioms is related to the comprehension process. As wesaw previously, some idiomatic phrases behave like big words and cannot bevaried either syntactically or lexically. Other idioms, however, are quite flexible(Wasow, Sag, & Nunberg, 1982). So we can say "after the lunch break the airwas cleared" as a syntactic variant of the idiom to clear the air. The degree ofsyntactic flexibility seems to be related strongly to the level of analyzability of anidiom (Gibbs & Nayak, 1989; Gibbs, Nayak, Bolton, & Keppel, 1989; Gibbs,Nayak, & Cutting, 1989); the more analyzable they are the more syntacticflexibility they exhibit. The levels of analyzability suggested by Cacciari andGlucksberg (1990) relate to the way in which the constituent parts of an idiommap onto their idiomatic referents. Idioms classified as analyzable and trans-parent have constituents that map directly onto their respective idiomatic refer-ents. For example, in the idiom pop the question, the verb pop maps to itsreferent "suddenly ask" and the noun phrase "the question" maps to the "mar-


riage proposal." This contrasts with the classification analyzable and opaque,where the relation between an idiomatic constituent and its referent is not seman-tic, but rather is in some sense indirect. For instance, in the idiom spill the beans,there is no direct semantic relation between beans and "secrets." Finally, idiomsclassified as nonanalyzable possess constituents that do not map onto the constit-uents of the idiom's referent. The constituent words kick and bucket in the idiomkick the bucket, for example, bear no relation to the idiomatic meaning "to die."Henceforth in this chapter, we refer to this relationship between the constituentwords of an idiom and the constituents of the idiomatic meaning as the referentmapping.

We want to argue that idiom comprehension involves the recognition of thereferent mapping associated the idiom and that the structure of the mappingdetermines the degree of syntactic and semantic flexibility the phrase exhibits.The differences is analyzability reflect the different types of data structuresmapped by an idiom's referent mapping. For nonanalyzable idioms the underly-ing referent mapping merely takes word strings into word strings and no furtherstructure is transferred. For analyzable idioms, on the other hand, the referentmapping takes constituent words into constituent words but in a structured man-ner. For instance, in our example pop the question the verb and object of theidiom map to the verb and object of its meaning, respectively. In other words, thereferent mapping preserves the linguistic functional structure of the idiomaticphrase. The mapping can be thought of as carrying information from the surfaceform of an idiomatic phrase to the "deep" or semantic representation of thephrase. The additional information encoded in a syntactic variant of an idiom canbe carried to the underlying meaning representation only if the correspondingreferent mapping preserves it.

For a nonanalyzable idiom like kick the bucket the range of syntactic variationis minimal, allowing differences in tense such as kicks the bucket and kicked thebucket, but not passivization, "the bucket was kicked." The latter transformationwould involve a restructuring of the word string, which is not permitted by theidiom's string-to-string referent mapping. Variations in tense are allowable be-cause tense can be regarded as a global feature of the word string and in thissense is external to it. Information about the tense might be detected at thesurface level by a morphemic analyzer and passed independently of the idiomaticword string to the underlying meaning structure. String-to-string referent map-pings realize the idioms-as-big-words hypothesis in that the internal structure ofthe mapped strings is opaque to them. The referent mappings of analyzableidioms, on the other hand, map the linguistic functional structure of the strings.

In this approach it is necessary to distinguish two levels of linguistic represen-tation involved in idiom comprehension. The first level is the surface representa-tion of the idiomatic phrase encoded as a constituent tree. In the system wedescribe, this level of representation corresponds to the constituent structure built

235


by the parser as it configures the semantic information associated with the con-stituent lexical elements. At this level, the parser is able to identify the wordstrings that comprise big word idioms.

The second level of representation encodes the predicate-argument structure,or functional structure, of the phrase. That is, how the main predicate of theidiom, generally the predicate associated with the verb, relates to its argumentssuch as the subject and object grammatical functions. This level of structure alsoencodes the predicate-argument structure of the idiom's underlying meaning. Inour system, these two structures are stored as part of the lexical entries associatedwith the constituent lexemes of the idiomatic phrase. The referent mapping,binding the two structures, is encoded as a set of substitutions that replace theconstituent elements with their corresponding referent elements in the semanticrepresentation of the idiom.

In terms of this representational framework, the referent mappings of non-analyzable idioms bind the surface level representation of the idiom directly to itsassociated meaning structure, whereas the corresponding mappings for analyza-ble idioms bind the idiom's predicate-argument structure to the equivalent struc-ture for its semantic representation. Analyzable idioms permit syntactic flexibil-ity because, under such variations, their representations remain unchanged at thelevel of predicate-argument structure. For example, the phrase "the beans werespilt by John" is interpretable because at the predicate-argument level of repre-sentation its structure, SPILL[subj(John), obj(the beans)], is the same as for thenonpassivized form of the phrase. Figure 10.1 depicts the referent mapping forthis phrase, showing how the predicate-argument structure of the phrase is pre-served by the substitutions that implement it. If the referent mapping did notdistinguish this level of structure, analyzable idioms could not be recognizedunder such transformations. For nonanalyzable idioms, the predicate-argumentstructure of the idiom's referent meaning is retrieved as a whole by the idiom'ssurface-level word string.

It should be noted that we are not attempting to explain the origins of thedifferent categories of referent mapping, that is, why the idiom kick the bucket isa word-string-to-word-string mapping, whereas for other idioms it preserves the

[RED tSPLL(subj obj RED ITELL(subj, obj)

SUBJ IJohnl susiOBJ Ihe beans - OBJ the secre1si

Predicate.argument structure Predicate-argument structureof the Idiomatic phrase of the idiom's referent meaning

FIG. 10.1 Referent mapping for the idiom to spill the beans.


linguistic structure. As we already have said, we believe these origins to beburied in the cultural history of a language. Instead, we want to claim that idiomcomprehension corresponds to the recognition of the referent mapping underly-ing an idiom and that such mappings are encoded in the lexicon. If this is thecase, then a computational model of idiom comprehension comprises a parsingmechanism for recognizing the referent mappings of idioms based on their lex-ical constituents.

PROCESSING IDIOMS

Before discussing our own approach to parsing idioms, we briefly mention someof the other approaches that have been taken. In some systems a simple pre-processor is dedicated to the recognition of idiomatic forms. This preprocessorreplaces the group of words that comprise the idiom with the word or words thatconvey its meaning. In Augmented Transition Network (ATN) systems, es-pecially if oriented toward a specific domain, particular sequences of arcs some-times are inserted in the network which, if transited, leads to the recognition of aparticular idiom (e.g., PLANES: Waltz, 1978). LIFER (Hendrix, 1977), one ofthe most successful applied systems, is based on a semantic grammar and withinthis mechanism, idiom recognition is implemented without considering the syn-tactic flexibility of idioms, that is, the possibility of changing the syntacticstructure of the idiom while still preserving its figurative meaning.

PHRAN (Wilensky & Arens, 1980) is a system based entirely on patternrecognition. An interesting aspect of PHRAN is that it has a twin generator calledPHRED (Jacobs, 1985) that reverses the process. PHRAN and PHRED share thesame database of patterns (including single words at the same level). PHRANnormally proceeds deterministically, applying the two principles "when in doubtchoose the more specific pattern" and "choose the longest pattern." The limits ofthis approach lie in its low capacity for generating various alternative interpreta-tions in the case of ambiguity (although a nondeterministic implementation alsowas realized eventually and in running the risk of having too great a spread ofnonterminal symbols (i.e., word's syntactic class) if the database of idioms islarge. Work on idioms that takes a similar perspective and that pays particularattention to the problem of learning idioms is that of Dyer and Zernik (1986).

A more recent approach to processing idioms is that of van der Linden (1991).His original approach was based on a connectionist model similar to the onedeveloped by Cottrell (1988) for the resolution of lexical ambiguity (van derLinden & Kraaij, 1990). The model consisted of four levels of units; the lowestlevel units encode form, the units on the next level encode syntactic discrimina-tions, which activate units on the next, semantic, feature level. Finally, thesemantic feature units activate relational nodes in the semantic network. Incommon with many connectionist models, excitatory links exist between levels

237


and inhibitory links within levels. The meanings of idioms are encoded in thesemantic network and a particular idiom is recognized when its unit emerges asthe strongest competitor at this level, inhibiting the various nonidiomatic read-ings. One problem with this approach that van der Linden (1991) himself hasidentified is that it is computationally expensive, requiring extensive parallelprocessing resources. A more serious drawback, and one that is shared with otherconnectionist natural language models, is that the network must encode all thepossible syntactic and semantic forms that a particular idiom might take. Suchmodels have difficulty when they encounter a new form for an idiom because theinformation necessary to distinguish it from possible nonidiomatic readings is notrepresented within the network and hence is unable to contribute in support of theidiom's semantic network unit. For big word idioms this is not much of aproblem, but for idioms that exhibit some degree of syntactic flexibility and foridioms in languages based on relatively free word order, this problem quicklybecomes intractable. This type of reasoning led van der Linden (1991) to movetoward a symbolic hierarchical model in which idioms are presented as part of alexicon incorporating an inheritance hierarchy, very much in line with the ap-proach we propose next.

Our approach is implemented in a system called WEDNESDAY 2 (Stock,1989)-a system whose design was intended to be to some extent psychologi-cally plausible in that it was inspired by empirical evidence from idiom com-prehension studies (Cacciari, 1988; Swinney & Cutler, 1979). It is intended to bea flexible approach designed to integrate idioms into the lexicon simply as moreinformation about particular words rather than as a separate list requiring specialtreatment as suggested by Katz (1973).

Because the processing of idioms in WEDNESDAY 2 exploits the basicmechanisms of the parser, it is helpful to understand the parser independently ofidiom processing in the first place.

WEDNESDAY 2 is a nondeterminisfic parser that can explore different alter-natives when in a given state and with a given data situation. So, on the onehand, it can commit itself to one particular choice, but on the other hand, iteventually might provide all of the possible interpretations of a sentence or of afragment. If working with finite resources, and in particular if the basic apparatusis sequential, the problem of managing partial structures and the control of theprocess becomes crucial.

Representation of Linguistic Knowledge

The representation of linguistic knowledge in WEDNESDAY 2 is based on twodistinct syntactic aspects often merged together in other approaches. One is theinformation carried by morphemes-information that is combined in word speci-fications. This includes local syntactic information and entities for which we usethe metaphor of "impulses." The role of these impulses in parsing is to provoke


the linking together of fragments of semantic information on the basis of syntac-tic constraints. Yet, technically speaking, they are expressed in a declarativeform and it is the parsing processor that makes use of them. The other syntacticaspect is constituted by restrictions on the space where this linking can occur.Maintaining the previous metaphor, this means limiting the space where im-pulses "operate." This latter aspect is responsible for the structuring of thesentence into constituents. What is crucial for this approach is that the positionsof elements inside a constituent and their functional aspects are not determinednecessarily by the second component but depend on the information included inthe lexicon. So it is possible to deal with a whole range of characterizations fromrigid word-order languages, to languages in which only dependencies betweenparticular parts of speech must hold, down to free word-order languages.

A word interpretation (there can be more than one interpretation for a word)includes a semantic interpretation and syntactic information of various kindsincluding subcategorization specifications that appear as alternative specifica-tions in impulses. So an impulse basically specifies all the ways in which a givenlinguistic function is to find its value. It also includes measures of likelihood thatare used to derive an overall measure of likelihood of a partial analysis. Aparticular indication (i.e., "main") is provided for the head element of theconstituent at the semantic level.

The only other data that the parser uses are recursive (nonaugmented) transi-tion networks that provide restrictions on search spaces where impulses can lookfor requested elements. In other words, these networks deal with the distributionof constituents. A particular kind of arc is designated to indicate that only theoccurrence of something expected by preceding words (i.e., for which an im-pulse was set up) will allow the transition to occur.

As far as the position of elements within WEDNESDAY 2 is concerned, onecan specify in a natural and nonredundant way all the gradations from obligatorypositions to obligatory priorities to simple likelihoods of relative positions.

The Parser

The parser builds on the idea of chart parsing (Kaplan, 1973; Kay, 1980).Previous nondeterministic parsers usually incorporated a backtracking mecha-nism so that when the analysis following the current alternative failed an earliercontext was restored and the process continued following one of the possiblealternatives in that context. Using backtracking in this way means that a fragmentmight have to be reanalyzed many times along different alternative paths. Chartparsing avoids all this; a chart parser is willing to adopt any strategy of analysis,but whatever the adopted strategy, work is never duplicated. Basically, a chartcan be seen as a working memory structure in which representations are kept forcomplete constituents, for incomplete constituents (together with the indicationof what is needed to complete them), and for working hypotheses. The chart is

239


built "around" the sentence and therefore both the relations of dominance andprecedence are represented within it. The basic entities of the chart are called"edges" (in fact, they indicate the portion of a string that they refer to). "Inactiveedges" denote completely recognized constituents whereas "active edges" referto constituents that have been recognized only partially. The process of parsing isbasically a process of introducing new edges into the chart in order to extend theexisting edges so as to include Jarger portions of the sentence. In WEDNESDAY2 active edges are used as search spaces in the sense just outlined. Actually, theyare complex data structures with rich in formation including a semantic interpreta-tion of the fragment, syntactic data, pending impulses, an overall measure oflikelihood, and so forth. Data on an edge are unified (merged together) dynam-ically as indicated later.

The mechanism is capable of processing different subproblems in parallel. Infact, working on a serial machine, a multiprocessing scheme is realized with anagenda that at every moment of the computation includes the list of tasks to becarried out. A scheduling function decides the priorities to give the various tasks.An interesting aspect is that heuristics easily can be included to guide the parser(Stock, 1989), thus implementing strategies that may take the context into ac-count: For instance, the semantic relevance of a certain partial interpretation canbe evaluated and can cause the selection of that particular task to be carried onfirst by the processor; or more structural criteria can be adopted such as, "Carryon the task that will extend most a partial analysis chosen amongst the largestactive edges present so far in the chart."

The parser works asymmetrically with respect to the "arrival" of the mainelement. Before the main element arrives the extension of an edge has almost noeffect. Upon the arrival of the main element, all the candidate fillers must find acompatible impulse and all impulses concerning the main element must findsatisfaction. If all this does not happen then the new edge expected to be added tothe chart is not added: The situation is recognized as a failure-an impossibletentative analysis. After the arrival of the main, each new head must find animpulse to merge with and each incoming impulse must find satisfaction. Again,if all this does not happen the new edge will not be added to the chart.

Mechanisms for handling certain forms of ellipsis and for treating long-distance dependencies such as those found in relative clauses also are included inthe system. These mechanisms again rely on lexical data and impulses on the onehand, and on constraints defined in the networks on the other.

The relevance of the chart-based approach can be appreciated best through anexample. Let us consider the parsing of the sentence, "The king gave the girlwith the long black hair to the sheikh." The fragment "the girl with the longblack hair" is analyzed only once, resulting in a noun phrase (NP), to which acertain semantic interpretation can be associated; this analysis is always avail-able, regardless of how it originated. For example, the lexical entry for the wordgave may have specified the alternatives that an NP following the verb can be


either the direct object (semantically, the transferred object) or the indirect object(semantically, the recipient). In the lexicon, the likelihoods of these situations arespecified; but in addition, the compatible situations for the other grammaticalfunctions involved with the word gave are specified. So if the first NP processedafter gave is to be considered the direct object, then a prepositional phrase (PP)marked to must appear subsequently in the sentence and semantically specify therecipient. Alternatively, if the first NP is to be considered the recipient, thetransferred object will follow, specified as an NP.

At the sentence level, the material "the girl with the long black hair" canoccupy two completely different roles. When the material is processed in com-bination with the previous partial analysis of the sentence, the parser establishestwo possible tasks that are inserted into the agenda. In principle they could beconsidered in parallel; if we have a single processor, one of the tasks is chosenfirst (say the one with the NP having the recipient interpretation, because of itssemantics). When it is noticed that the following PP ("to the sheikh") is notcompatible with this partial interpretation, the other task is executed, finding asalready completed and available the analysis of the fragment "the girl with thelong black hair." The final PP is compatible with the second interpretation andthe sentence is interpreted completely.

Specification of Idioms in the LexiconThe power of WEDNESDAY 2 idiom processing is bound to flexible idioms.Nonflexible idiom forms that are to be treated as big words constitute no problemfor any parser. Idioms as big words can be represented in the lexicon indexed bythe initial words of the string.

Flexible idioms are represented in WEDNESDAY 2's lexicon as further speci-fications of words just as in a dictionary. In this first realization, idioms arebound to the particular word that represents the crucial part of that idiom. InWEDNESDAY 2 terms, the word that represents the crucial part of the idiom isthe one that bears the main of the immediate constituent including the idiom. Forinstance in build castles in the air it would be the verb to build. The choice of thepresent realization can be extended to other elements of the idiomatic expressionand possibly to multiple entries of the same expression.

Each idiom is described in two sections: The first describes the elementscharacterizing that idiom, expressed coherently with the normal characterizationof the word. The second section describes the interpretation, that is, whichsubstitutions should be performed when the idiom is recognized.

Constituents of an idiom are described as particular fillers of linguistic func-tions or particular modifiers. Again using the example of build castles in the air,when build is in an active form it has castles as a further description of the fillerof the object (OBJ) function and the string in the air as a further specification of aparticular modifier that may be attached to the main element. In addition, there is

241


an indication of the weight that the particular idiom element carries for theoverall recognition. At the same time it always remains possible to specify a newalternative for a filler for an existing function including the description of thecomponent and its weight. For instance, a new alternative might be a partial NPinstead of a complete NP as in "take care" or an NP marked differently fromusual. Finally, the entry can include the information that a certain impulse spec-ified for the word is to be considered to have been removed for this idiomdescription (e.g., "go tell some one else" where the direct object of tell is absentwhereas in literal expressions it must be there).

Elements of an idiom can be specified in the lexicon as string patterns,semantic patterns, value restrictions ,and so forth), and morphological varia-tions. Coreferences among parts of an idiom (such as "John made up his mind,"where the mind is John's mind) also can be specified there. Substitutions to beperformed when the idiom is recognized include the semantics of the idiom thatare supposed to take the place of the literal semantics, plus a specification of thenew main and of the bindings for the new functions in relation to the "literal"ones. For analyzable idioms, substitutions are only possible for referent map-pings that map the main-functions (i.e., predicate-argument) structure of theidiomatic phrase into the new idiomatic representations. In the case of non-analyzable idioms, the semantics of the idiom must be substituted as a whole unitbecause the lexical entries for the idiom's constituent lexemes do not containidiomatic readings that preserve the rnain-function information relaxing to theinternal structure of the phrase.

Idiom Processing in WEDNESDAY 2

Idiom processing in WEDNESDAY 2 is integrated into the nondeterministicmultiprocessing-based behavior of the parser. As the literal analysis proceeds andpartial representations are built, the background impulses are monitored, check-ing for possible idiomatic fragments. Monitoring is carried out only for frag-ments of idioms not in conflict with the present configuration. A dynamic activa-tion table is introduced with the occurrence of a word that has some idiomspecification associated. The weight associated with each constituent of an idiomexpresses the relevance of the constituent in the recognition of that idiom; so theoccurrence of an expected fragment of an idiom gives strength to that hypothesisin proportion to the relative weight of the fragment.

If the configuration of the sentence conflicts with one element then the idiomthat includes that element is discarded from the table. For example, with thesentence "John built a house for his parents," after a house is interpreted as thedirect object of built, the idiomatic entry associated with build castles in the air isdiscarded from the activation table. The normal processing goes on, includingthe possible nondeterministic choices, the establishment of new processes, andso on. The activation tables are included in the edges of the chart.

10. SOME COMPUTATIONAL AND THEORETICAL ISSUES 243

The weights associated with the different constituents of (many of) the idiomsrepresented in WEDNESDAY 2 were taken from the results of an empirical studyof idiom comprehension conducted by Cacciari (1988). Cacciari created six listsof (Italian) idioms, each comprised of 13 idioms and 40 literal sentences asfillers. Each list contained a fragment of each idiom and of a filler in a random,prespecified order. The length of the fragments varied in the six lists, so that inthe first list the fragments were the shortest (e.g., "He was . . .") and in the listsixth list, the longest (e.g., "He was sitting on pins and .). Each list wassubmitted to a different group of 15 subjects in such a way as to ensure that nosubject saw the same idiom twice. Subjects were presented with incompletesentences written in a booklet and were asked to complete them in a meaningfulway. The percentage of subjects correctly completing an idiom with the differentamount of information given in the six lists was used as the basis for determiningthe weight of the constituents of the 80 idioms constituting the corpus.

When the activation level of a particular idiom crosses its associated threshold(corresponding to the recognition point mentioned previously), another process isintroduced dedicated solely to that particular idiom. The function of this processis to build the idiomatic interpretation. The process begins by introducing anedge in which substitutions are carried out that are compatible with all theinformation derived by the analysis up to that point. The process continues aslong as the incoming information is consistent with the idiomatic representationit is building. The original process goes on as well (unless the fragment givingrise to the new process is not syntactic, being peculiar to that idiom alone); theonly change being that the idiom is removed from the active idiom table. At thispoint there are two working processes and it is a matter for the (external) schedul-ing function to decide on the priorities. That is, according to different processingmodels one can decide at that point to assign priority to the literal interpretationor to the idiomatic interpretation.

But still the processing is open and the idiomatic process may result in failure:Further analysis may not confirm what has been hypothesized as an idiom.Furthermore, a different idiomatic process may be separated from the literalprocess at a later stage when its own activation level crosses the threshold.

As an example, we consider the idiom to build castles in the air in the contextof the (defective) sentence, "The tycoon built castles in the Highlands." At somepoint, an attempt at an idiomatic analysis of the sentence will be undertaken andfail at which point it will be replaced by the literal analysis:

1. When built is analyzed, its idiom activation table is instantiated.

2. When castles is interpreted as the direct object of built, the activation ofthe idiom referred to previously crosses the threshold as specified in the lexicon.An additional (idiomatic) process starts at this point with the new interpretationunified with the previous interpretation of the subject: The process just departsfrom the literal process; no backtracking is performed. At this point we have two


processes going on: an idiomatic process in which the interpretation is already"the tycoon created unrealistic plans," and the literal process that, of course,incorporates the background monitoring for other potential idioms that mightfeasibly be consistent up to that point: of the analysis.

3. The modifier "in the Highlands" is recognized; the idiomatic process failsbecause the spatial modifier is different from the one expected.

4. The literal process yields its analysis.

The flexibility inherent in the WEDNESDAY 2 analysis of idioms is requiredto accommodate the phenomenon of word-order flexibility that is present in alanguage such as Italian, even if it is of minimal importance in English. Forinstance, in the Italian version of the same idiom, namely costruire castelli inaria, it is acceptable to have a form like "Mario costruf nella sua vita diversicastelli in aria" (word by word: "Mario built in his life several castles in the air")and even, with sarcasm in the topicalization: "In aria Mario costruf diversicastelli" ("In the air Mario built several castles"). Although in English this idiommight be processed as a big word, it is clear from the examples that this cannot bethe case in Italian.

EXTENSIONS AND PROBLEMS

In principle, this view can be extended quite easily by linking an idiom descrip-tion not just to one entry (e.g., the verb for sentences) or verb-phrase-(VP) levelidioms) that "monitors" the recognition, but to more entries that refer to thesame description (such as the NPs or P'Ps involved). With this refinement, themost meaningful component of an idiom will instantiate the table. The sameprocess would be established and the same representation could be used. Theimportant aspect here is that idioms are recognized by means of the literalanalysis and therefore the process makes uses of the syntactic productivity and ofthe syntactic constraints available. Van der Linden (1991) proposed a hierarchicalorganization of the lexicon for specifying idioms. We think this is very much inthe spirit of our proposal and may lead to a better kind of factoring of commonparts of larger idioms.

So, this is a framework in which one could insert data relative to the access ofidioms and a theory about how idioms might be processed. This framework iscompatible with relevant experimental results. For example, it allows for thesimultaneous accessibility of the literal meanings of the component words of anidiom during comprehension. Cacciari and Tabossi (1988) provided direct evi-dence for such activation during idiom comprehension. Using a cross-modalpriming paradigm they showed that the literal meanings of the component wordsof an idiomatic expression are activated immediately, but that the idiomaticmeaning is not activated until 300 ins later, when both the literal and idiomatic


senses are active. Even in studies using contexts that are appropriate to theidiomatic meanings of the phrases, both the idiomatic and literal meanings aredirectly accessible. Such data are consistent with the organization of a parser thataccesses both the idiomatic and literal lexical entries for the input words and isable to construct both literal and figurative interpretations of the input in parallel.The actual distribution of processing resources over the two interpretations isdetermined by various higher level factors such as immediate semantic anddiscourse constraints.

Dead metaphors could fit well into this picture. Through semantic interpreta-tion of the autonomous parts (going on in the literal interpretation), in particularas far as modifiers are introduced, one could import something into the newidiomatic/metaphoric meaning. For instance, in the current setting of WEDNES-DAY 2 a modifier of the head of the idiom (like "John kicked the bucket 2months after the accident") is correctly interpreted as modifying the idiomaticmeaning ("John died 2 months after the accident"); the same holds for mor-phological modifiers (like "Jack rehit the road"); and there is also the possibilityof using material that modifies a nonverbal component in the idiomatic in-terpretation. For instance, one could specify that material that modifies a nouncan be used to modify the "verb predicate" of one of its arguments in theidiomatic interpretation. A case of this kind could be "Mary built professionalcastles in the air," meaning that she had unrealistic professional fantasies. Novelmetaphors, however, require a different mechanism because they involve specifi-cations at the knowledge level.

The framework we are proposing incorporates the recognition of idiomaticreference as an integral component of natural language processing, but at thesame time accommodates the syntactic flexibility exhibited by idioms in mostlanguages. However, no attempt is made to capture other forms of productivity,as we believe them to be the product of processes independent of the recognitionprocess.

REFERENCES

Cacciari, C. (1988). La comprensione delle espressioni idiomatiche: 11 rapporto fra significatoletterale e significato figurato [The comprehension of idiomatic expressions: The relation-ship between literal and figurative meanings]. Unpublished doctoral dissertation, University ofBologna, Italy.



Clark, H. H., & Lucy, P. (1975). Understanding what is meant from what is said: A study inconversationally conveyed requests. Journal of Verbal Learning and Verbal Behavior, 14, 56-72.

245


Cottrell, G. W. (1988). A model of lexical access of ambiguous words. In S. L. Small, G. K.Cottrell, & M. Tanenhaus (Eds.), Lexical ambiguity resolution. San Mateo, CA: Kaufmann.

Dyer, M., & Zemik, U. (1986). Encoding and acquiring meaning for figurative phrases. In Pro-ceedings of the 24th Meeting of the Association for Computational Linguistics (pp. 106- 111).New York.

Gibbs, R. W. (1983). Do people always process the literal meanings of indirect requests? Journal ofExperimental Psychology: Learning, Memory and Cognition, 9, 524-533.

Gibbs, R. W., & Nayak, N. (1989). Psycholinguistic studies on the syntactic behavior of idioms.Cognitive Psychology, 21, 100-138.

Gibbs, R. W., Nayak, N., Bolton, J. L., & Keppel, M. (1989). Speakers' assumption about thelexical flexibility of idioms. Memory and Cognition, 17, 58-68.

Gibbs, R. W., Nayak, N., & Cutting, C. (1989). How to kick the bucket and not decompose.Analyzability and idiom processing. Journal of Memory and Language, 28, 576-593.

Gibbs, R. W., & O'Brien, J. (1990). Idioms and mental imager: The metaphorical motivation foridiomatic meaning. Cognition, 36, 35-68.

Glucksberg, S. (1990). Beyond literal meanings The psychology of allusion. Paper presented at theannual meeting of the American Psychological Association, Boston.

Hendrix, G. G. (1977). LIFER: A natural language interface facility. SIGARTNewsletter, No. 61.Jacobs, P. (1985). PHRED: A generator for natural language interfaces. Computational Linguistics,

11(4), 219-242.Kaplan, R. (1973). A general syntactic processor. In R. Rustin (Ed.), Natural language processing.

Englewood Cliffs, NJ: Prentice-Hall.Katz, J. (1973). Compositionality, idiomaticity, and lexical substitution. In S. R. Anderson & P.

Kiparsky (Eds.), A festschrift for Morris Halte. New York: Holt, Rinehart & Winston.

Kay, M. (1980). Algorithm schemata and data structures in syntactic processing. Xerox Palo AltoResearch Center, Palo Alto, CA.

Lakoff, G. (in press). Metaphor: The contemporary view. In A. Ortony (Ed.), Metaphor andthought (2nd ed.). New York: Cambridge University Press.

Nayak, N., and Gibbs, R. W. (1990). Conceptual knowledge in the interpretation of idioms. Journalof Experimental Psychology: General, 119, :315-330.

Ortony, A. (Ed.). (in press). Metaphor and thought (2nd ed.). New York: Cambridge UniversityPress.

Ortony, A., Schallert, D. L., Reynolds, R. E., & Antos, S. J. (1978). Interpreting metaphors andidioms: Some effects of context on comprehension. Journal of Verbal Learning and VerbalBehavior, 17, 465-477.

Stock, 0. (1989). Parsing with flexibility, dynamic strategies, and idioms in mind. ComputationalLinguistics, 15, 1.


Van der Linden, E. (1991). Idioms, non-literal language and knowledge representation. Proceed-ings of the lJCAI workshop on computational approaches to non-literal language: Metaphor,metonymy, idiom, speech acts, implicature (pp. 141-150). Sydney, Australia: Conference Orga-nization.

Van der Linden, E., & Kraaij, W. (1990). Ambiguity resolution and the retrieval of idioms: Twoapproaches. Proceedings of COLING-90 (Vol 2, pp. 1-6). Helsinki, Finland: Conference Orga-nization.

Waltz, D. (1978). An English language question answering system for a large relational database.Communications of the Association for Computing Machinery, 21(7), 526-539.

Wasow, T., Sag, 1. A., & Nunberg, G. (1982). Idioms: An interim report. Proceedings of the XIIIInternational Congress of Linguistics (pp. 102--105). Tokyo: CIPL.

10. SOME COMPUTATIONAL AND THEORETICAL ISSUES 247

Wilensky, R., & Arens, Y. (1980). PHRAN-A knowledge-based approach to natural languageanalysis (ERL Memorandum No. UCB/ERL M80/34). Berkeley: University of California, Elec-tronic Research Laboratory.

Wood, M. W. (1986). A definition of idiom. Bloomington: Indiana University Linguistics Club.

The Recognition andInterpretation of Idioms

Stephen G. PulmanUniversity of Cambridge Computer Laboratory andSRI International Cambridge Computer Science Research Centre

Here are some examples of phrases that have been described as idiomatic in boththe traditional and the generative linguistic literature:

1. John turned on the light.2. Joe put up with the noise.3. Bill took advantage of her generosity.4. He did it for his mother's sake.

5. The old curmudgeon finally kicked the bucket.6. We got him to spill the beans.7. That really put the cat among the pigeons.

8. He was chasing a red herring.

9. We feel they are skating on thin ice with that project.

10. He shouldn't have counted his chickens before they were hatched.

Intuitively speaking, these examples are ranged on a scale from the more gram-maticalized, like turn on, to phrases that legitimately might be regarded asproverbs rather than idioms, like the count your chickens example. In betweenare examples like spill the beans, or skate on thin ice. What they all have incommon is that their meanings are not deducible from the ordinary meanings oftheir components by the usual rules of compositional semantics. We do not getthe meaning of turn on in this context from a combination of the meanings ofturn and on, and although the meanings of skate, thin, and ice surely play somekind of part in our interpretation of Sentence 9, the example nevertheless conveys

249

250 PULMAN

something over and above a picture of people moving rapidly across ice onskates.

To say that idioms are not interpreted compositionally is not intended to beclaiming that noncompositionality is a sufficient condition for an idiom. Nor is itto deny that there is no trace of compositionality in these examples. Not every-thing that is noncompositional normally would be counted as an idiom. Forexample, a phrase like on top of, which corresponds to no productive composi-tional semantic pattern (*on bottom of), is probably best thought of as a "com-pound preposition." And for X's sake is recognizably the same idiom as for thesake of X and, with the exception of the contribution of sake, is as compositionalas possessives in general are. Nevertheless, idioms are not wholly composi-tional.

This lack of total compositionality may be for one of several reasons: Perhapsthe components do not in isolation have a meaning distinct from their occurrencein the idiom, as with sake infor X's sake, or tabs in the idiom keep tabs on; or itmay be that they have several meanings (or one radically underdetermined one),the correct one of which can be determined only in the context of the idiom, aswith up and with and on and so forth in the first two examples. More characteris-tically, perhaps, it may be that the components of the idiom have literal mean-ings, but that these are not what is involved in their interpretation as an idiom.This latter type of idiom often is characterized (metaphorically) as a frozen ordead metaphor. However, its meaning is not completely lexicalized: It is cer-tainly the case that someone unfamiliar with the idiom nevertheless can arrive atan appropriate meaning for it by processing it as a metaphor (Examples 7, 9, and10). As one might expect if one is treating it as a metaphor, one sometimesarrives at rational interpretations that are nevertheless not that of the idiom; forexample, detecting connotations of cruelty in the cat among the pigeons exam-ple: There is no such connotation in its ordinary usage.

Idioms are pervasive in all styles of language use. The problem they present tothe theoretical and computational linguist is not the fact that their meaning cannotbe worked out by the usual syntactic and semantic rules, for if it were not forother factors this could be overcome by treating them as multiword lexical itemsto be looked up in a list in a fairly straightforward way. (A treatment like this isall that is necessary for things like on top of). The problem is rather that unlike(most) lexical items, (most) idioms have considerable internal structure thatseems to interact with the usual productive syntactic and semantic mechanisms ofa language in ways that render a "look-up" approach impossible in the generalcase. It is as if idioms have some of the properties of lexical items, in that someaspect of their meaning simply has to be looked up in a dictionary, or somerepository of idiosyncratic information, but also have some properties of regularsyntactic and semantic structures, in particular a fair amount of variability in theactual forms in which the idiom can be encountered: a problem for look-up-basedapproaches. Thus a grammatical theory that does not allow for this interaction

11. THE RECOGNITION AND INTERPRETATION OF IDIOMS

between conflicting sets of properties is likely to be found wanting in crucialrespects: The ability of a theory to accommodate the behavior of idioms has beenseen therefore as a major argument in its favor, whether in Standard Theorytransformational grammar (Fraser, 1970; Katz, 1973), Generative Semantics(Newmeyer, 1972), more recent transformational theories (Chomsky, 1980),Lexical-Functional Grammar (Bresnan, 1982), Generalized Phrase StructureGrammar (Gazdar, Klein, Pullum, & Sag, 1985), or Tree Adjoining Grammar(Abeille, 1990).

The main point of the present article is to establish that there are someproperties of idioms that are not captured satisfactorily by any of the currentgrammatical frameworks, and that these properties suggest that idioms are in factnot properly to be considered as part of the domain of a grammatical theory at all.Instead, the recognition and interpretation of idioms belongs in the relativelyunexplored areas between compositional syntax and semantics on the one hand,and contextually interpreted, fully resolved, interpretations of utterances on theother. Getting from compositionally derived meaning syntactic and semanticrepresentations to fully resolved contextual interpretations involves what can becalled contextual reasoning. This is the process of taking the information that canbe derived linguistically from a sentence and fleshing it out with informationsupplied by the local context or general background knowledge, leading to afully specific interpretation of that utterance of the sentence in that particularcontext. Contextual reasoning includes the various processes that establish thereferents of pronouns and other anaphoric devices, the resolution of ellipsis, andthe filling in of contextually dependent "vague relations" such as that associatedwith possessives, compound nominals, and verbs like have or do: In order tointerpret them correctly, one needs to know that the have of "John has measles"is not quite the same as the have in "John has a sister."

PROPERTIES OF IDIOMS

First let us review some of the properties of idioms that have figured in recentdiscussions of them (Bresnan, 1982; Chomsky, 1980; Gazdar et al., 1985;Wasow, Sag, and Nunberg, 1982). The first observation is an old one: that inpragmatically appropriate contexts some idioms can appear in syntactically vari-ant forms:

11. John turned the light on.12. Her generosity was taken advantage of.13. Advantage was taken of her generosity.14. He did it for the sake of his mother.15. The beans were spilled.

251

252 PULMAN

16. It was thin ice that they were skating on.17. Chickens that are counted before they are hatched aren't a good basis for

a bank loan.

But some idioms seem fairly resistant to this kind of paraphrase:

18. ?*The bucket was finally kicked by the old curmudgeon.19. *It was among the pigeons that he put the cat.

20. *He was chasing a herring that was red.

The only interpretation these sentences easily receive is a literal one. It has beenclaimed (Chomsky, 1980) that there is a difference between syntactic construc-tions in this respect: that, in particular, most idioms do not appear in tough-movement variants of basic forms even though other transformational para-phrases of them are permitted:

21. She is easy to take advantage of.22. *Advantage is easy to take of her.

(Tough-movement is the term used informally to describe the relation betweenthe approximate paraphrases "it is easy to please John" and "John is easy toplease," associated with a group of adjectives that include easy, tough, anddifficult.) However, it seems likely (,see following examples) that this is not asyntactic matter so much as one of finding the appropriate context in which the"focusing" effect of tough-movement can seem natural:

23. She's playing hard to get.24. That's a course which is proving hard to steer.

25. That lie is difficult to nail.

A second important observation about the properties of idioms is that they arein the main capable of "internal modification":

26. I have a theoretical axe to grind.27. He cut through a lot of red tape.28. They were skating on very thin ice.

Here it is the components of the idioms that are being modified: Example 26means "I have a theoretical point to make," not "Theoretically, I have a point tomake" (on the most natural reading of the latter, at least). Likewise, it is theamount of red tape that is being quantified, not the action of cutting through redtape. Furthermore, these modifications feel entirely natural and straightforward:

11. THE RECOGNITION AND INTERPRETATION OF IDIOMS 253

There is no sense of "playing with words," humorous intent, or metalinguisticcomment. Contrast these in both respects with the following examples:

29. He kicked the proverbial bucket.30. I'll keep a close eye on his progress.31. That turned out to be a very red herring.

Kick the proverbial bucket is clearly metalinguistic; in the next example, al-though close syntactically modifies eye, it is felt semantically to modify thewhole phrase: It means "I'll watch his progress closely." In a very red herringmost people feel that there is an element of play or humor involved in arriving atthe interpretation, although in order to do this some modification of a componentof the idiom is involved: We could paraphrase the result as something like "Thatwas an unusually irrelevant diversion." Now, most people would agree that todeal adequately with examples like these three examples, we have to go outsidegrammar to some theory dealing with the stylistic properties of language. But forthe previous set of examples, where there is no feeling that anything other thanthe normal processes of syntactic and semantic modification are at work, theimplication is that we must assign independent meanings to the components ofthe idioms in order that there is something there for the modifiers to modify.

This conclusion is further reinforced by the observation that, contrary to thetraditional generative wisdom (e.g., Bresnan, 1982, quoting Grimshaw), parts ofidioms can serve as antecedents to pronouns and ellipsis:

32. He turned the tables on me and then I turned them on him.33. They said the tide would turn, and eventually it did.

This possibility seems to correlate with that of internal modification: An idiomthat cannot be modified internally cannot provide anaphoric antecedents:

34. *I'll keep an eye on him and one on her too.

The inference from these observations is that the components that serve asantecedents must be individually meaningful, as usually is assumed in the casesnot involving idioms.

A further observation that we can add in support of this general conclusion isthat some idioms can be recognized, albeit in something of a stylistically markedway, even when incomplete:

35. That's a case of counting your chickens (before they are hatched).

36. This is just smoke without fire (there's no smoke without fire).

254 PULMAN

37. That suggestion came from the bottom of the barrel (cf. scraping thebottom of the barrel).

For examples like at least the 36 and 37, to assemble the correct interpretationrequires the components to be individually meaningful.

CANONICAL FORM THEORIES

Until recently, most linguistic theories, to the extent that they have tried tocapture the properties of idioms at all, have adopted one version or another ofwhat I call a canonical form approach, Starting from the point of view that theidiosyncratic nature of idioms is a lexical property in some sense, and that thisidiosyncrasy should be stated only once, rather than having to be repeated for allthe variant forms in which the idiom may appear, these theories adopt some levelof linguistic representation provided by the theory as the canonical level for thisstatement. That is to say, this level provides a normal form to which all differentoccurrences of the idiom can be reduced: The properties of the idiom then can bestated at this level once and for all, and the normal mechanisms provided by thetheory for syntactic variation will ensure that alternative variant forms of theidiom are related in some systematic way to the canonical form, allowing theproperties of all the different forms to be deduced.

For example, Chomsky (1980) proposed that idioms should be catered for byusing a special class of lexical rules operating at the level of "D-structure," alevel of underlying grammatical representation at which semantically importantproperties are exhibited via a canonical syntactic form. Chomsky's lexical idiomrules are sensitive to the presence of particular groups of lexical items andoperate so as to reanalyze, or provide an alternative analysis for, independentlygenerated structures. Thus structures like:

38. John [VP [V kicked][NP the bucket]], and

39. John [VP [V took][NP advantage][P ofjlNP Bill]],

will be reanalyzed lexically as:

40. John [VP (V kicked the bucketli, and41. John [VP [V took advantage of]INP Bill]],

with the idiomatic meanings being assigned as those of the newly created com-plex V: kick the bucket = "die," take advantage of = "exploit," roughly. Theoriginal syntactic structures are retained, though, allowing for the observedsyntactic variation. The inability of some idioms to undergo certain rules can bedescribed via a mechanism for marking the output of the idiom rules with a list of


transformational processes that are allowed or disallowed. Chomsky did not gointo any formal detail on the properties of these idiom rules but it seems fair toconclude on the basis of his examples that they have at least the followingproperties: (a) They operate on constituents, not arbitrary sections of tree; and (b)there is only one rule per idiom. The first property is required if the proposedmechanism for handling the semantics is to be viable: Idioms are "like lexicalitems"-that is, they are single semantic units. The second is not requiredlogically but seems to be the intent of all canonical form approaches: The point ofstating facts about the idiom at a canonical level of representation is that it allowsonly I statement of the facts about the idiom.

Bresnan (1982) proposed a mechanism with some similar features, but herethe canonical form is a level of lexical representation. Idiomatic meanings, sheposited, are to be stated in the lexicon, associated with the main functor of theidiom: Thus one sense of keep, that which appears in the idiom keep tabs on willbe as follows:

keep: V, keep-tabs-on ((subj),(on obj)); (obj form)=tabs

The verb will have a special meaning, and will be marked syntactically asrequiring the other components of the idiom: on followed by an NP consisting ofthe head N tabs. Alternative forms, such as the passive version, will be derivedvia the normal mechanism of lexical redundancy rules, just as for other transitiveverbs:

4 passive a

keep: V, passive, keep-tabs-on ((by obj/null), (on subj); (subj form)=tabs

These redundancy rules allow variant lexical entries to be deduced from asingle basic one, where the variant forms are predictable. They fulfill many ofthe functions served by lexically governed transformational rules in earlier theo-ries. The interaction of such rules suffices to derive all the valid alternativeforms: Exceptions presumably can be marked as such. Bresnan's (1982) ap-proach, given the properties of lexical redundancy rules, seems to share, mutatismutandis both of the properties (a) and (b) of Chomsky's (1980) approach. It is afirm prediction of both theories that all the alternative variant forms in which aparticular idiom can appear will be relatable to each other through the syntacticor lexical mechanisms of the grammatical theory: that all the variants can be"derived" from the single canonical form at which the properties of the idiom arestated.

Both of these theories have their problems. Firstly, as Gazdar et al. (1985)pointed out, it is not clear how a theory that regards idioms as essentiallysynonymous with atomic lexical items can assign to them enough internal struc-

255

256 PULMAN

ture to allow for internal modification and the apparent appearance of some oftheir components as antecedents. Likewise, our ability to recognize partial orincomplete idioms remains unexplained on such a view.

Second, it seems that the predictions of these theories about the variety ofidioms are false. Sets of sentences like these are (correctly) not related by regularsyntactic or lexical processes in any current theory of grammar known to me:

42a. He turned the tables on me

42b. The tables have turned.

43a. Now he's let the cat out of the bag.43b. Now the cat's out of the bag.44a. I've got some loose ends to be tied up.

44b. I'm tying up a few loose ends.44c. A few loose ends need tying up.45a. He really put the cat among the pigeons.

45b. Now the cat's well and truly among the pigeons.

Nevertheless, in these groups of examples, it is intuitively obvious that it isthe same idiom appearing in each sentence, rather than several separate, acciden-tally related idioms. However, this fact is not something that can even be de-scribed in the canonical form theories, for there is no canonical form at which theproperties of the idiom can be represented. At best, there will have to be severaldifferent statements for the various syntactic forms in which the idiom appears.Although workable, this is a clumsy and unattractive position to be forced intotaking.

As well as there being no syntactic canonical form shared by examples likethese, it is also true that the components of the idiom, in some cases, are notgrouped together under the same constituent. There are other examples of thesame thing, where the idiomatic component of a sentence consists, roughlyspeaking, of the subject and verb, but where the complement can vary:

46. The ceiling fell in on Pedro.47. The earth finally moved for Eniestina.48. Fortune smiled on Bonzo.

In Tree Adjoining Grammar (TAG), as described in Abeille (1990), a lexicalhead simply is associated with the different syntactic constructs ("elementarytrees") it can appear in. In the case of idioms, these are regarded as having"multicomponent" heads, like kick, the, and bucket, and they are associateddirectly with a set of elementary trees representing the variant syntactic forms theidioms can appear in. In contrast to the preceding theories, there is no require-


ment that such families of elementary trees be "natural" or predictable in thesense that, say, the tough-movement predicates or the passivisable verbs wouldbe assigned to a natural set of elementary trees. Other than the fact that they havedistributed heads, there is apparently no important distinction made betweenidiomatic constructions and nonidiomatic constructions. Although not strictlyspeaking a canonical form theory, TAG does make the claim that the mechanismsof syntactic variation for idioms are the same for nonidioms.

TAG fares somewhat better than the other two theories in accounting for manyproperties of idioms, but suffers the same problems with respect to those caseswhere not all of the idiom is present. In those cases no elementary tree will matchthe entry for the idiom and it will not be recognized.

All these theories fare badly on examples of idioms where there is lexicalvariation in one component, such as:

49. Put/lay/spread your cards on the table.

TAG has the same problems as the other theories in capturing the fact that theseare versions of the same idiom, intuitively, not three separate idioms with identi-cal meanings.

THE PARTIAL FUNCTION APPROACH

A second type of approach to the properties of idioms is taken by both Wasow etal. (1983) and Gazdar et al. (1985). Their proposals are couched within a mono-stratal theory of syntax: There is only level of syntactic analysis, and thussyntactic or lexical canonical forms are not an option. This approach to syntacticdescription (which of course has been enormously influential) is wedded to asemantic theory that is basically that of Montague (1974). Semantic interpreta-tion consists of translation into a (theoretically dispensable) logical form-anexpression of a formally defined artificial language called Intensional Logic-which has a recursively defined model-theoretic interpretation. In general themeaning of a constituent is either a function or an argument (which itself may bea function). Meanings of complex constituents are built up by applying themeaning of some of the components to the meanings of the others, as function toargument. The denotations (logicians' proxies for meanings) of these functionsare built up from a basic ontology of individuals, possible worlds, and settheoretic constructions based on these.

Normally speaking, the interpretations of these various functions, and thepredicates and constants of Intensional Logic, are total functions: that is, they aredefined for every object in the domain of the interpretation. Thus (to simplify)the interpretation of something like table might be a (function from possibleworlds to a) function from objects to truth values: a function that tells you, for

257

258 PULMAN

every object, whether or not it is a table. Mathematically speaking, we also canhave "partial functions"; functions that aie defined for some arguments, but notothers. A simple example of such a function might be division: division of X byY is defined for all values of X and Y except where Y = zero. In these cases thevalue of divide (X, Y) = undefined. Thus. division is a function that is defined forsome numbers, but not all (all except zero, in fact).

The partial function approach to idioms advocated by Gazdar et al. (1985) isas follows: We assign to each of the words appearing in the idioms, as well as itsusual literal meaning, if it has one, an idiomatic meaning. This is a partialfunction that is defined for some arguments and not others. It is defined, in fact,only for the arguments corresponding to t he other components of the idiom. Thusin the case where all the components of the idiom appear in an appropriateconfiguration, we will end up with a meaning that results from the combinationof all the components with each other. If some of the components are missing,we will have something that is not fully defined, that is, not meaningful.

This can be illustrated informally as follows: We assign to turn its usualmeaning, which we represent as turn and also an idiomatic meaning, roughlycorresponding to seize. We treat the other components of an idiom in a similarway. To suggest (to us) that the idiomatic components all belong together, weassign them all a numerical subscript. This also serves to distinguish the idi-omatic meanings from their literal counterparts: Turning the tables is not literally(i.e., physically) to seize anything, but it is to seize2! something. Now, the basicidea is that the grammar will assign to a sentence containing the words inquestion analyses corresponding to all possible combinations of the entries forthe words. However, of these combinations, only those that have all the literal orall the idiomatic senses (in our inforrnal illustration, those with the same sub-script), combined in the appropriate function-argument pattern, will have a prop-erly defined meaning:

turn: turn, seize 2,the: the, a2 , . . .

tables: tables, position-of-strength 2 1 . -turn the tables =(i) turn (the (tables))(ii) seize21 (a2, (position-of-strength 2 d)(iii) *turn (a2, (position-of-strength 2 l))(iv) *seized (the (tables))etc.

Combinations of literal with idiomatic senses are not defined: They do notcombine into a sensible meaning.

Because according to this view the components of idioms are independentlymeaningful, it is not surprising that they can be modified, can serve as anteced-ents, and so on. (Those idioms that do not have these properties are treated


essentially as complex lexical items by the Generalized Phrase Structure Gram-mar (GPSG) account: They thus distinguish two subtypes of idioms).

There are, unfortunately, one or two problems with the partial function ac-count. One question concerns the basis on which meanings are categorized intoliteral and nonliteral. An intuition behind the GPSG treatment (I am grateful toGerald Gazdar for making me aware of this) was that the mechanism could beregarded as an extension of the notion of a "sorted logic." In a sorted logic, theobjects in the domain are separated into sorts, and the various expressions of thelogic are categorized as to what sort of thing they can be predicated of mean-ingfully. To give an informal example, one might say that the predicate drink inEnglish was (ignoring metaphorical uses) required to have something of sort"living thing" as its subject and sort "liquid" as its object. It is not meaningful tocombine drink with arguments of differing sorts: "The book drank the flower"sounds odd.

In the case of idioms, the effect of this mechanism is to make a combinationlike "*seize21 (the (tables))" as anomalous as "the book drank the flowers,"and for essentially the same reason. But if the motivation behind the assignmentof things to sorts is semantic, it is not obvious that the restrictions on combina-tion that are required actually could be obtained in a way that ensures overallconsistency of the sortal system. For example, if we assume that the idiomaticmeaning of axe to grind is essentially that of point to make, then we might assignthe sort of point to the idiomatic sense of axe, and that of make to grind.However, there then would be no obvious reason why I cannot say "I want toconsider this axe" with the meaning "I want to consider this point." If considercan combine nonidiomatically with point, it is not obvious why it cannot com-bine with the relevant meaning of axe without sortal violation.

If we carry out the assignment of partial meanings on some other basis, thereare still problems in ensuring that only the right combinations are permitted. Theinitial assumption is that literal verbs will have to be defined in such a way thatthey do not combine with idiomatic arguments, for otherwise we would be ableto say things like "seize the tables," meaning "seize a position of strength." Thissuggests that the sortal structures erected for idiomatic expressions should beinsulated from those that apply to nonidiomatic expressions, although parallel tothat one in various respects. However, things are not quite so simple, for at leastsome literal verbs do seem to combine with idiomatic components, provided therest of the idiom is there too:

50. He tried to break the ice which inhibited our conversation.

In order to account for the interpretation of the relative clause, in which the ice isthe subject of inhibit (via its connection to which), we would need to assume, itseems, a special sense of inhibit that is capable of combining with idiomaticsenses.

Gazdar et al. (1985) apparently were willing to acknowledge this multiplica-

259

260 PULMAN

tion of word senses. They pointed out that partial functions will "permeate theentire lexicon" (p. 239), even though this proliferation apparently leads to theprediction that combinations like:

51. The ice inhibited our conversation.

Should be easily interpretable, without contextual priming, and with the idi-omatic components receiving the interpretations they would have in the fullidiom. My intuition is that this is not the case. However, even if it is, there surelywill be a considerable combinatorial explosion problem if this program is carriedthrough systematically: Just consider a phrase like see the light. This will havemany different analyses: a literal one, an idiomatic one involving the specialsense of see required to combine with the light of the light at the end of thetunnel; a second idiomatic one in which a complete idiom is present (meaningsomething like "suddenly have the truth revealed"); all possible combinations ofthese, resulting in structures with no fully defined meaning; combinations ofeach of these with analyses in which the the is the the of kick the bucket (nb.*kick a bucket), or of cat among the pigeons (nb. *cat among no pigeons),likewise yielding no fully defined meaning, and so on and so forth.

Nor is this multiplication of meanings restricted to the nonlogical vocabularyof the meaning representation. It is also the case that all the logical constants ofIntensional Logic will have to be made systematically ambiguous between totaland partial function uses. For example, if the partialness of idiom denotations isto be made to do the work it is required to do, it must be inherited by functionsapplying to those partial functions. This will happen straightforwardly in mostinstances, by the normal compositional rules of the grammar. But in some cases,the components of an idiom are in structures where they are not operating on eachother as function to argument, being combined in some other structure. Considerthe case of:

52. That has put a cat among the pigeons.53. Now there's a cat among the pigeons.

Assuming that the relevant parts of the idiom are a cat and among the pigeons,then the rules assumed for these constructions by Gazdar et al. (1985) willdeliver structures like these (ignoring intensional operators for simplicity) for thetwo sentences:

54. Put (among (the pigeons))) (a cat)).55. Be (a [Ax. cat(x) A [among (the (pigeons))](x)]).

In the first of this pair, put combines with among the pigeons to form a predicatethat can apply to a cat. Put will pass on the partialness of its argument. But in thesecond, there is no verb that applies to among the pigeons and a cat in a similar


way. Thus in order to rule out combinations of literal cats with idiomatic pigeons,and suchlike oddities, it seems that it must be arranged that and is defined in sucha way that it only accepts two idiomatic arguments or two literal arguments, andnot combinations. This is technically possible, of course (though the resultingadjustments to the model theory of Intensional Logic are not trivial), but isbeginning to seem a somewhat less than elegant solution. It is, of course, also anapproach that will present some serious practical problems if one is interested inbeing able to produce a computationally feasible analysis of idioms: It in effectmakes the problem of automatic word-sense selection many times more complexthan it was before. For now, many sentences that do not obviously containambiguous words will do so under this analysis simply because of the fact thatthey contain some words that are a component of some idiom, the other compo-nents of which are not actually present in the current sentence.

IDIOMS AND QUASI-INFERENCE

It seems to me that the way out of the inadequacies and problems of the preced-ing theories is to acknowledge and make use of the observation that in order forcertain types of idiom to be regarded as having a canonical form, either of thesyntactic-lexical type envisaged by Chomsky (1980) and Bresnan (1982), or"purely semantic," which is what the Gazdar et al. (1985) proposal amounts to,then we must give up the assumption that this level of canonical form corre-sponds to anything made available in current theories of grammar as an indepen-dently motivated level of representation. More specifically, the relationship be-tween pairs like the two examples just discussed, or the following is not anythingthat could or should be captured by syntactic or lexical rules, or directly bycompositional semantics:

56. John let the cat out of the bag.

57. The cat is out of the bag.58. He turned the tables on me.59. The tables have turned.

60. He laid his cards on the table.

61. His cards are on the table.

If anything, what we have here is a relationship of entailment. In each case thefirst member of the pair entails the second. In the case where we are dealing withthe more usual examples of idioms appearing in related constructions, the samething is true: Characteristically, what makes us decide that two syntactic con-structions are related is that sentences in one form entail sentences in the otherform, and sometimes vice versa.

If this observation is correct, then several consequences follow: First, if we

261

262 PULMAN

want to state the facts about an idiom just once-if we regard pairs of sentencelike Examples 56-61 as involving the "same" idiom-then that statement willhave to be at some postgrammatical level of representation. Second, it followsfrom the nature of some of these entailments that the properties of the idiom willhave to be stated on the entailed member of the pairs in the aforementionedexamples, not the entailing member. That is to say, the idiom must be thecat . . . out of the bag, a cat . . . among the pigeons, and not let the cat out ofthe bag, or put a cat among the pigeons.

Let us see how such a proposal might work out in detail. We first of all outlinea purely mechanical solution, which is capable of being implemented in a prac-tical natural language understanding system, and then go on to discuss thetheoretical status of the constructs involved.

We assume some grammatical theory providing translations of sentence mean-ings into logical forms directly. Thus if we envisage such a grammar as used by aparser, the parser will deliver, for a given input string, one or more logical forms,if the input string was covered by the grammar. These logical forms will feed intoan inference mechanism of some kind. This inference mechanism also will haveaccess to various meaning postulates governing the behavior of all the nonlogicalconstants of the logic that we are translating sentences into (at least as manyconstants as content words in the language, let us assume). Among these will bepostulates giving information about the inferences arising from causal verbs likelet, put, lay, and so on, as follows:

62. Vx,P.let(x,P) -- cause(x,P).63. Vx,yz. put(x,yz) -> location(v,z).

We now add to these a set of "idiom rules" somewhat similar in form butdescribing the relation between literal and idiomatic interpretations:

64. Vx,y. cat(x) A bag(y) A out-offx,y) z 3a,z. secret(z) A revealed(a, z).65. Vx,p. cat(c) A pigeons(p) A location(c,p) 3y.disturbance(y).

The application of these rules is triggered by the appearance of a logical form that"matches" the left-hand side of an idiom rule. When this is the case, the idi-omatic meaning is a possible one for the sentence as well as, or perhaps insteadof, the literal meaning, depending on the context.

From:John let the cat out of the baglet(John,3c,b. cat(c) A bag(b) A out-of(c,b))via:3c,b. cat(c) A bag(b) A out-of(c,b)and:


Vx,y. cat(x) A bag(y) A out-of(x,y) ~ 3a,z. secret(z) A revealed(a, z)to:3a,z. secret(z) A revealed(a, z)

From:John put a cat among the pigeons3c,p. cat(c) A pigeons(p) A put(john,c.p)via:3c,p. cat(c) A pigeons(p) A location(c,p)and:Vx,y. cat(x) A pigeons(y) A location(x,y) 3z.disturbance(z)to:3z.disturbance(z)

We can treat, purely mechanically, the process of matching idiom rules anddrawing out the idiomatic meaning as ordinary proof theoretic inference operat-ing on the logical forms produced by a natural language system for some giveninput sentence. We just treat the connective as if it were -A, take the logicalform representing the literal sentence meaning (perhaps transformed into somekind of normal form), and by a process of forward chaining inference on it usethe idiom rules to produce all the derivable "conclusions." Among these will bethe idiomatic interpretation of the sentence "deduced" by the idiom rules.

Using something that has the properties of inference has several advantages.So, for example, the ordering of conjuncts in the literal logical form does notmatter; nor does the presence of extra conjuncts reflecting the appearance ofinternal modifiers of components of the idiom. Treating the process as akin toinference provides the necessary flexibility to deal with these types of variation,whereas a simple "pattern-matching" approach would have problems charac-terizing the range of permissible variant forms from the target pattern. A simplepattern-matching approach would require, for example, that all of the inputs thatshould trigger a particular idiom rule had the same overall pattern: But of coursethis cannot be guaranteed to be the case easily if there are many variant syntacticforms of an idiom. Partial idioms and modifying phrases also would producestructures not necessarily matching the patterns provided. Inference, on the otherhand, simply requires that variant forms logically entail the "antecedent" of therule: They do not have to be syntactically similar, much less identical.

Under this quasi-inference scheme, then, the sequence of events in analyzingthe purely compositional meanings of a sentence, in a naive "classical" naturallanguage processing system will be: (a) Parse it, (b) apply compositional seman-tic rules to give the set of literal logical forms for the sentence, and (c) use therelevant set of meaning postulates and idiom rules to expand the set of interpreta-tions.

This set of literal and idiomatic interpretations then can be checked against the

263

264 PULMAN

context to discover what the most plausible meaning of the sentence is on thisparticular occasion.

However, it is very important to notice that our idiom rules are neither ordi-nary meaning postulates nor inference rules. They are not meaning postulates,because they do not restrict the meanings of the literal interpretations on theirleft-hand side. The literal meaning is what it is and is not changed by the fact thatan idiomatic reading is also possible. Nor are they inference rules, because theserequire it to be the case that their antecedent is true before their consequent canbe regarded as true: Inference is the process of going from true premises to trueconclusions. But in the case of the rules giving idiomatic interpretations, inalmost all but a few unusual circumstances it is precisely the fact that theantecedent is not literally true that makes the idiomatic interpretation plausible.

This makes it rather important to discover exactly what these idiom rules are.One possible interpretation, suggested to me by Robin Cooper (personal commu-nication), of the status of idiom rules is as follows. Think of the rules citedearlier, of the general form literal idiom, as being a quantification oversituations, meaning roughly, "in any situation in which idiom would be literallytrue, it is permissible to use literal to describe it." This is a plausible view,although it seems to me to give idioms more of a metalinguistic status than onemight have thought they deserved.

A more simple rationalization for the idiom rules might be along somewhatsimilar lines, but without the metalinguistic flavor or the vocabulary of situationtheory of the previous suggestion. The ordinary definition of truth for a logicallanguage like that we have been assuming might supply a statement of the truthconditions for a logical form like:

66. 3c,b. cat(c) A bag(b) A out-of(c,b).

This would be in terms of something like the following: ". . . is true if and onlyif there is some a cat c and a bag b and the pair (c,b) is a member of thedenotation of the predicate out-of. Such descriptions of truth conditions forlogical forms are deduced in a recursive way from statements of truth conditionsfor the basic predicates, constants, and logical devices of the language. What wedo now is to think of idioms as simply adding extra clauses to this set of basicdefinitions, to the effect that a logical form like the aforementioned also will havethe truth conditions: ". . . is true if and only if there is some secret s, and there issome agent, a, and the pair (a,s) is a member of the denotation of reveal.

Of course, the mechanism for recursively assigning truth conditions will haveto operate in a way analogous to our quasi-inference described earlier, if it is tobe able to recognize when these extra clauses can apply. But this picture of theprocess of interpreting sentences is not completely unfamiliar. Many aspects ofthe interpretation of compositionally derived logical forms involve a processsyntactically indistinguishable from inference but without requiring the semantic


warrant usually demanded of "real" inference. For example, one needs to make"deductions" about possible resolutions of ellipsis, pronoun reference, and soforth, in order to arrive at a fully fledged logical form that can be assessed fortruth and falsity. A sentence as such does not contain all the information neces-sary to build a proposition: Some must be supplied by the context. The processby which this is done, in many computational approaches at least, involvessomething that looks like reasoning, but with hypotheses rather than establishedfacts (Alshawi, 1992). (Some of the inferences do involve facts, too, of course).The general term for the types of reasoning involved in this aspect of computa-tional linguistics is contextual reasoning. The application of idiom rules shouldbe thought of as simply another instance of this kind of reasoning, in whichquestions about the truth of the propositions involved is suspended. (There mightbe a connection here with the "methodological solipsism" of Fodor, 1981). Thereasoning simply is cranked out on a purely syntactic basis and it is not untilthose wheels have finished turning that the results are evaluated for truth.

On the assumption that this picture is unproblematic, then the status of ouridiom rules also becomes unproblematic: They just operate to give an extra set oftruth conditions to sentences interpretable as idioms. For some idiomatic sen-tences, the usual mechanism for assigning truth conditions in fact will fail: Forexample, "John has a large sake" should receive no truth conditions, becausesake is meaningless outside of the idiom it usually occurs in.

FALSE POSITIVES

This mechanism for the interpretation and recognition of idioms enables us to dowithout special senses for the words involved in an idiom. The idiom rules can bedefined in terms of the constants representing the literal meanings of the wordspresent. This avoids the technical problems raised earlier, as well as the potentialcombinatorial explosion of literal-idiom hybrids caused by the partial functionapproach. All the words, or rather, their translations into logical constants, mustbe present, and they must combine in an appropriate way to produce a particularlogical form for the idiom to be recognized. (Notice that this presupposes a fairlyfine-grained view of the relation between lexical items and logical constants: thatthe word sense is pretty much the primitive element of meaning, no furtherdecomposition into "semantic atoms" or suchlike being required. This view isargued for in some detail in Pulman, 1983).

However, using literal senses of words to trigger the recognition of idiomsbrings some potential problems. We might find some idioms accidentally,through the use of other lexical items connected to those triggering the idiomrules via meaning postulates. Take the following example:

67. We watched the chandelier being switched on.

266 PULMAN

This might be taken to entail:

68. We saw the light.

This particular example has an idiomatic reading. But the former does not havethe idiomatic meaning associated with the latter.

In some cases, though, we can find "inferred idioms," as in the followingexample:

69. John put the tabby among the pigeons.

Then, assume the following is given:

70. Vx. tabby (x) -- cat (x).

This in turn will entail:

71. John put the cat among the pigeons.

Then, the idiomatic quasi-inference will go through.One in fact does hear circumlocutionary versions of common idioms for

humorous effect ("then the manure really entered the ventilation system"), and itis possible, given an appropriate setting, to interpret "put the tabby among thepigeons" along similar lines. Nevertheless, there is often a conscious piece ofreasoning to do in order to arrive at these interpretations, with associated im-plicatures generated. I take it that the normal process of interpreting idioms doesnot involve such conscious reasoning, and, accordingly, that we have to solve theproblem of "false positives."

It may be that in deriving idiomatic interpretations from a level of meaningrepresentation at which the original lexical items are no longer present, we havegone too far. Nevertheless, we do not want to make too much depend on actuallexical items, because of the fact that we discussed earlier: Some idioms canallow any of a range of semantically similar words to occur in a particularposition.

We could retain some connection between the actual lexical items used and theidiom rules by developing some kind of indexing scheme for idiom rules basedon the occurrence of particular lexical items in the input sentence. Thus, anysentence mentioning, say, cats and bags might cause the relevant inference ruleto be triggered. On sentences like the following, the rule would be invoked, butwould not succeed:

72. The cat meowed as her owner emptied the fish out of her shopping bag.


On sentences like this one, however, the idiom rule would not be invoked, eventhough the sentence might entail a cat being out of a bag:

73. The tabby emerged from the sack with a mouthful of salmon.

When a sentence is processed, only those idiom rules suggested by the indexingmechanism are used in trying to find extra idiomatic readings. The fact that someidioms display some lexical variation then can be accommodated, because pre-sumably the varying item is not crucial for the presence of the idiom.

With a scheme like this, the circumlocutionary idioms mentioned earliershould be recognized only when this mechanism has been overridden: Their self-conscious nature suggests that a fair degree of problem solving may be involvedin recognizing them. One other benefit of a mechanism for lexical indexing isthat it would enable us to account for the possibility of recognizing incompleteidioms. If a particular sentence makes no literal sense, but the lexical itemspresent narrow down the range of possible idioms to a manageable number, thenthe retrieval of the appropriate idiomatic meaning still will be possible. If recog-nition was only by the idiom rules, then nothing less than the complete idiomwould suffice for its successful recognition.

Here then is a simple scheme for lexical indexing. We associate an identifierwith each distinct idiom rule, and for each such identifier we decide on some setof lexical items (in their morphological base form) that we take as a cue for thepresence of that idiom:

{skate,on,thinice} = idiomn31{break,the,ice} = idiom-256{cat,out,of,bag} - idiom-45{cat,among,pigeons} = idiom-42etc.

Now, when in the process of working out the interpretations for a sentence, wetake the set of lexical items in it, L, and for each of the aforementioned sets ofwords, S, if S is a subset of L then the idiom may be present. Notice that thisdoes not tell us whether the idiom is present: For that, we have to go though allthe mechanisms outlined earlier. All that this indexing does is to tell us whichidiom rules in principle can apply to this sentence: Whether they actually do ornot will depend on the logical form that the sentence has. For example:

74. John was skating on thin ice. (idiom possible and rule fires)75. They wanted to skate but the ice was too thin. (idiom possible but rule

fails)76. The ice was thin. (idiom not possible)

268 PULMAN

In the case where we have an idiom fragment, then, in the absence of anysatisfactory literal meaning we can perform the aforementioned indexing testusing "intersection" rather than "subset" to generate candidate idiom rules. Thiswill tell us that if there is no other interpretation for Example 76, the idiomaticone is a possibility, although some more powerful mechanism than the ones wehave developed will be necessary to produce it.

SYNTACTIC VARIATION

A further consequence of the quasi-inferential treatment of idioms is that therecan be no stated connection between an idiom and the various syntactic configu-rations that it can or cannot appear in, at least on the assumption that syntacticproperties are not encoded into logical form and thus available for inference.Thus any syntactic variant that produces the same logical form should be capableof giving rise to an idiomatic inference. This is apparently not so, at least forsome examples. However, I think it is possible to explain away this apparentproblem for our account for this along lines suggested partly by Newmeyer(1972) and partly by Gazdar et al. (1985).

Consider what the pragmatic function of syntactic variation is. Presumably apassive, or a clefted, or a topicalized fon-n of a sentence is chosen, not because aproposition expressed in this way cannot be expressed any other way, but becausethe demands of the discourse context are such that the information contained inthe proposition must be presented in a certain way. At a simple level, this muchcan be demonstrated by the relative oddity of discourses like the following:

77. Speaker 1: Who hit Bill?78. Speaker 2: It was Bill who was hit by John.

Although Speaker 2 expresses a proposition that gives an answer to Speaker l'squestion, he does it in an odd way. The emphasis is on Bill, whereas theemphasis of the question was on whoever it was that hit Bill. On the assumption,then, that syntactic forms are going to sound odd if there is no context in whichthe way they present their information is plausible, we can explain the oddity ofmany syntactic forms of idioms as follows: For a syntactic form to be appropri-ate, there must be a context in which it makes sense to focus or contrast thediscourse entity that the construction does focus or contrast. In the case wherethere is no such discourse entity, oddity will result. Thus, roughly speaking, thefollowing sounds odd because it focuses on the bucket:

79. *The bucket was kicked by John.


But the bucket does not correspond to any discourse entity, and thus there is nocontext in which this will be appropriate. In other words, as suggested in adifferent framework by Newmeyer (1972), the sentence sounds odd for much thesame reason that the following sounds odd:

80. *John was died.

However, in an idiom like spill the beans, it is possible to associate (idiomatic)discourse entities with the components of the idiom:

81. Vx,y. beans (y) A spill(x,y) -* 3i,a. information(i) & reveal (a, i).

Thus it is possible to find a context in which it makes sense to focus one or theother:

82. The beans were spilled.

This sounds fine, roughly because the following sounds fine:

83. The information was revealed.

Even cases where in isolation a syntactic variant sounds peculiar, the idiomaticmeaning can be rescued by providing a context in which it makes sense toemphasize the discourse entity that is being focused. Consider the idiom drop abrick, meaning "make a stupid or embarrassing mistake." It is very difficult torecover the idiomatic interpretation from a single sentence like:

84. What John dropped was a brick.

This is because the pseudocleft construction is emphasizing the discourse entityreferred to by brick and in a null context no such entity is available. But incontexts where the pragmatic function of the pseudocleft makes sense, the idi-omatic interpretation is perfectly accessible.

We need a context in which it makes sense to give extra emphasis to anidiomatic brick that has been dropped. Fortunately there is an idiom that gives usthe right contrast: To drop a bombshell means to reveal some surprising orshocking information. So we can now contrive a context in which two idiomaticthings are dropped:

When he revealed the company finances,John thought he'd dropped a bombshell,but what he'd really dropped was a brick.

270 PULMAN

I find this occurrence of the pseudocleft perfectly natural with the idiomaticmeaning, in contrast to the earlier uncontextualized one.

It thus seems legitimate to regard all idioms as being able in principle to occurin any syntactic configuration, and therefore we can put the responsibility forexplaining why some variations sound better than others on to some future theoryof information structuring in relation to syntax.

ACKNOWLEDGMENTS

This chapter arose from a talk given to a conference on figurative uses oflanguage organized by Bar-Ilan University, Israel, in 1986. I am grateful to theBritish Council for funding my trip there, to Ellen Spolsky for making it pos-sible, and to the participants at the conference itself for their good-humoredguidance around a culturally (in several senses) unknown landscape. Subsequentversions of the article were given to the Linguistics Association of Great Britain,in Edinburgh, and in talks to the Moral Sciences Club of the University ofCambridge, and to linguistics departments in Essex, Manchester (UMIST), andSussex. I am grateful to participants on all these occasions for their comments,and in particular to Hiyan Alshawi, Robin Cooper, Gerald Gazdar, Malka Rap-poport, Susan Rothstein, and Graham Russell.

REFERENCES

Abeille, A. (1990). Lexical and syntactic rules in a tree adjoining grammar. In Proceedings of 28thannual meeting, Association for Computational Linguistics (pp. 292-298). Pittsburgh: ACL.

Alshawi, H. (1992). The core language engine. Cambridge, MA: MIT Press.Bresnan, J. (1982). The passive in lexical theory. In J. Bresnan (Ed.), The mental representation of

grammatical relations (pp. 40-65). Cambridge. MA: MIT Press.Chomsky, N. (1980). Rules and representations. Oxford, England: Basil Blackwell.Fodor, J. A. (1981). Representations: Philosophical essays on thefoundations of cognitive science.

Sussex, England: Harvester.Fraser, B. (1970). Idioms within a transformational grammar Foundations of Language, 6, 22-42.Gazdar, G., Klein, E., Pullum, G. K., & Sag, 1. (1985). Generalised phrase structure grammar.

Oxford, England: Basil Blackwell.

Montague, R. (1974). Formal philosophy. New Haven, CT: Yale University Press.Newmeyer, F. (1972). The insertion of idioms In P. M. Pergateau, J. N. Levi, & C. Phares (Eds.),

(pp. 282-295). Papers from the 8th regional meeting. Chicago Linguistics Society, Chicago, IL.Pulman, S. G. (1983). Word meaning and belief. London: Croom Helm, and Norwood, NJ: Ablex.Wasow, T., Sag, I., & Nunberg, G. (1982). Idioms: An interim report. In Preprints of the plenary

session papers, X1I1th International Congress oj Linguists (pp. 87-96). Tokyo: CIPL.

1The Determiner inEnglish Idioms

Christiane FellbaumPrinceton University

Idioms have been the subject of investigation by linguists and psychologists for anumber of years. Both groups have been concerned largely with the representa-tion and status of idioms in the lexicon, that is, with the question as to whetheridioms are stored, accessed, and subject to grammatical rules in the same way assingle lexical items. Whereas the experiments of psychologists by and large haveinvestigated how speakers process nonliteral language, linguists generally havetried to determine the status of idioms in the lexicon by comparing their syntacticbehavior with that of single-word lexical items. The central question is whetheror not idioms can be shown to have a meaningful internal structure, that is,whether or not they are decomposable into individual chunks or constituents thatare semantically nonvacuous (Nunberg, 1978; Pulman, 1986; Wasow, Sag, &Nunberg, 1983; and many others).

Contrary to earlier findings, the general consensus among researchers in thefield now is that many idioms are compositional. These idioms are distinguishedfrom noncompositional ones where the meanings of the individual lexical com-ponents are opaque, that is, where such components cannot be assigned eitherliteral or figurative interpretations that contribute to the meaning of the entireidiom. Examples of noncompositional idioms are trip the light fantastic and kickthe bucket, which are unanalyzable except as wholes, and thus function seman-tically as a single word. (Like compositional idioms, these strings are subject tocertain low-level syntactic operations such as subject-verb agreement.)

By contrast, speakers treat compositional idioms as analyzable strings. Evi-dence for this comes largely from the syntactic operations that idioms toleratecompared with nonfigurative strings; for example, passivization commonly hasbeen taken as a test for the semantic transparency of nouns in idioms. Idioms

271

272 FELLBAUM

vary with respect to the degree of syntactic and lexical frozenness, and theirflexibility appears to be straightforwardly correlated with their compositionality(Gibbs & Nayak, 1989; Gibbs, Nayak. Bolton, & Keppel, 1989; Newmeyer,1974; Wasow et al., 1983); Nunberg's (1978) important distinction betweennormally and abnormally decomposable idioms accounts for some of the syntac-tic and semantic behavior of many idioms.

Our work follows this line of investigation by showing how the nature of thedeterminer points to the semantic status of noun phrases (NPs) in many idioms,which in turn predicts the range of possible determiner variations.

The syntactic operations (Fraser, 1970; Ruwet, 1973; and others) and lexicalsubstitutions (Gibbs & Nayak, 1989; Gibbs et al., 1989; and others) that haveserved to test the robustness of the idiom's figurative meaning always haveinvolved content words, usually nouns and verbs. No attention has been paid tothe status of the determiner apart from the noun that it precedes in assessing thesyntactic or lexical frozenness of idioms. Although it is undoubtedly the contentwords that make an idiom a special kind of lexical unit, we argue that, in manycases, the determiner carries a considerable semantic load; its shape in a givenidiom contributes to the interpretation of the noun and hence to the particularquality of figurative language. Like nouns and verbs, determiners are poly-semous in that a given determiner can precede nouns with different kinds ofsemantic status.

Determiners behave like nouns and verbs in that some idioms tolerate theiralterations or substitutions whereas others do not. Although a strong correlationcan be shown to exist, the flexibility of the determiner in an idiom is notnecessarily a predictor of the idiom's overall syntactic or semantic flexibility; forexample, an idiom whose indefinite determiner can be exchanged for a quantifierdoes not necessarily passivize.

Flexibility of the determiner is found only in compositional idioms, where it islargely an indicator of the noun's referential status. In the large class of noncom-positional, unanalyzable idioms like trip the lightfantastic and buy thefarm, thedeterminer is invariable. Moreover, the forms of the determiners here cannot beexplained in terms of the rules for literal language that appear to operate in manyof the compositional idioms. However, we suggest that the definite article that isfound in most of these strings has a particular function relating to the figurativenature of these strings.

PROPERTIES OF THE DETERMINERIN IDIOMS

Sentences 1-3 show that, like in literal language, different kinds of determinersand nondescriptive adjectives occur with the NPs in verb phrase (VP) idioms

12. THE DETERMINER IN ENGLISH IDIOMS 273

(this study will be limited to VP idioms of the form [V][NPJ[PP], where V =

verb and PP = prepositional phrase):

1. Torture can make you spill the beans. (definite)2. Mary will take a shot at the bar exam. (indefinite)

3. The minister resigned in order to save face. (zero)

I refer to the expressions in Sentences 1-3 as idioms in their dictionary forms,and I assume that these citation forms correspond to the context-neutral represen-tations in which idioms are stored in the mental lexicon of speakers. The deter-miners that occur in the dictionary forms of idioms constitute only a subset of thedeterminers available in literal language. For example, it is hard to find idiomswhose VP-internal nouns in their dictionary forms are preceded by general quan-tifiers such as every and many, or by a negative determiner. Similarly, no demon-strative or specificity-inducing adjectives seem to occur in idioms listed in dictio-naries. However, as Sentences 4-7 show, idioms can be used with suchdeterminers and adjectives without a resultant loss in the figurative reading:

4. Mary had several shots at the exam, but always failed.

5. Mike has no bone to pick with Joe.6. I knew all the chairmen of the committees, but I thought it too risky to pull

those strings.7. John had a specific axe to grind with Bob.

The fact that idioms can preserve their figurative meaning when the determiner ischanged contradicts the assumption that they are represented as single lexicalitems in the mental lexicon of native speakers (Bobrow & Bell, 1973; Swinney &Cutler, 1979; and others). The determiner changes in Sentences 4-7 occur insidethe idiom, showing that its individual constituents are subject to syntactic andand semantic operations. This supports the now common contention that idiomsdo not differ in many ways from "literal" strings. The data in Sentences 4-7confirm the observations made by a number of researchers that many othersyntactic and semantic alterations do not result necessarily in the loss of thefigurative reading of idioms. More important, because the determiner, as well theadjectives in Sentences 6 and 7, reflect the referential status of the nouns,the possible changes give important clues to the nouns' semantics.

The substitution of a quantifier in Sentence 4 and a negative in Sentence 5constitute semantic alterations, in that they change the meanings or truth valuesof the sentences. These changes are analogous to Gibbs et al.'s (1989) substitu-tion of semantically related nouns and verbs in VP idioms and Nunberg's (1978)insertion of descriptive adjectives. The specificity-inducing adjective in Sentence

274 FELLBAUM

7 distinguishes this noun from, for example, shot in the idiom have a shot at,which does not tolerate the presence of specific. This indicates that the role ofaxe and shot within the VPs somehow must differ, despite their identical dictio-nary forms.

The use of a demonstrative adjective, as in Sentence 6, shows that the idiomcan be altered in accordance with pragmatic discourse requirements, that is, thatthe lexical material inside the VP is discourse sensitive: The use of the demon-strative is licensed by the reference to the preceding material. I

However, although there is considerable flexibility in the determiners of idi-oms, this flexibility is not unconstrained. Like other lexical items or strings,idioms can be manipulated syntactically and semantically only in accordancewith speakers' grammatical competence and the pragmatic conditions in a givendiscourse. These rules of language use constrain the possible grammatical con-figurations of both literal and figurative language. But in addition to the con-straints that apply to idioms as grammatical strings there are constraints whoseviolations, although not affecting a sentence's grammaticality, will result in theloss of the phrase's figurative meaning. Thus, Wasow et al. (1983) pointed outthat speakers know whether a given VP idiom can or cannot be passivized,without ever having been exposed to the relevant data. Furthermore, asyntacticidioms like kiss ass should not be available for the rules governing literal lan-guage because they do not match the input strings for those rules; yet, their NPscan be modified. This knowledge, too, is part of a speaker's linguistic knowl-edge.

If idioms were represented in speaker's lexicons as individual items, bothsyntactic idioms and such asyntactic strings like kiss ass and kiss a lot of asswould have to be acquired and stored separately. However, it seems more plausi-ble that the NP has its own semantic representation, which determines, accordingto the same rules as in literal language, which changes are tolerated by the NPand which ones result in a loss of the figurative meaning. We show that thesemantics of many idiom NPs decide the range of permissible determiners.

Although most NPs in literal language are compatible with several differentdeterminers (subject to the semantics of the noun and discourse-pragmatic fac-tors), the determiner in many idiom NPs can be varied to a lesser extent; speakersknow that whereas the VPs in Sentences 4-7 preserve the idiomatic reading, thesentences that follow are to be interpreted either literally or as idiosyncraticvariations by a humorous speaker:

8. Mary had shots at the exam but always failed.9. 1 thought it too risky to pull the strings.

'We are not concerned here with NP-internal syntactic rules like the one requiring agreement ofthe possessive adjective with the subject applying in the many body-part idioms like put one's footdown: John finally put his foot down.


10. John spilled no beans.11. Max had the axe to grind with Bob.

In other idioms, the determiner tolerates no alternation; thus, Sentence 13 doesnot receive the figurative interpretation of Sentence 12:

12. You will have to bite the bullet.13. You will have to bite {a/no/every} bullet.

The nonarbitrariness of the determiner indicates that it is an important part of theidiom. Its behavior with respect to insertion, deletion, and substitutability resem-bles that of the content word constituents of most idioms. We examine thesystematic constraints on the semantic, syntactic, and pragmatic uses of deter-miners in idioms. We see that, in many idioms, the nature of the determinerfollows from the semantics of the NP, and that its behavior can be accounted foralong the same principles that hold for NPs in the literal language.

THE LEXICAL REPRESENTATIONAND VARIABILITY OF IDIOMS

The lexicon of a language can be thought of as a mapping of meanings, orconcepts, and words, which are the lexicalized realizations of these concepts(Saussure's signifies and significant, respectively). We refer to words that arelinked with a meaning in this way as "denoting." NPs in idioms, when they aredenoting, are just like words receiving a literal interpretation in that theyare linked to certain specific meanings. These meanings may or may not beexpressible by other, nonfigurative nouns. For example, the sense of ice in theidiom break the ice cannot be referred to easily by a single English noun; only theword ice lexicalizes the relevant concept including tension, shyness, stiffness,uneasiness in a group.

Denoting nouns in idioms resemble conventional metaphors, which are nounsused figuratively independent of a certain circumscribed idiomatic context. Bothmetaphors and denoting nouns occurring in idioms are represented multiply inthe mental lexicon; that is, they are polysemous. Thus, the noun bean(s) in thecontext of the VP idiom spill the beans is associated with roughly the samemeaning as the words secret and (confidential) information; bacon in bring homethe bacon is interpreted in the same way as financial support, earnings, income.The difference between the sense of beans that is linked to the meaning legume orvegetable and the sense of beans that is linked to the meaning secret is that thelatter interpretation is generally available only in the context of the VP idiom spillthe beans, whereas the former can occur in many more contexts that are less

276 FELLBAUM

constrained. Standard dictionaries, such as Collins, reflect this by including inthe entries for spill and bean, besides their nonfigurative meanings, the idiomaticphrase as a whole.

Another way to look at the lexicorn is to think of the word form bean asassociated with (at least) two different concepts or meanings, one being"legume," the other, roughly, "secret.'

Many meanings are associated with morc than one word, giving rise to thefamiliar phenomenon of synonymity. Although in one of its senses, the wordbeans is a close synonym of legume, in its other sense it is a synonym of secret.In the literal language, synonyms may be substituted more or less freely for eachother without a resulting change in the meaning or truth value of the phrase orsentence. In idioms, however, substituting synonyms of either the literal or thefigurative meaning often results in the phrase or sentence receiving a literalinterpretation only. Thus, Sentence 14 is very likely to be interpreted literally,and Sentence 15 is interpretable only as a speech error or a pun:

14. He spilled the legumes.

15. He spilled the secret.

Some idioms have slightly varying dictionary forms. Thus, Boatner, Gates, andMakkai (1975) list both button one's lips and zip one's lips; here, two closelyrelated, though non synonymous verbs, can appear in the same idiom. But we donot find substitutions of other, similarly related verbs such as tielclip/knot/velcroone's lips.

Gibbs et al. (1989) replaced verbs and nouns in idioms with semanticallysimilar words and asked subjects to rate the meaning similarity between thealtered phrases and the original idioms. They found that subjects judged manyidioms with lexical substitutions such as burst the question to be semanticallyclose to their canonical forms; in this case, pop the question. In general, Gibbs etal. found that subjects rated lexical alterations in semantically normally decom-posable idioms less disruptive to the idioms' meaning than in abnormally decom-posable and nondecomposable idioms. Thus, subjects stated that they found thesimilarity between phrases like punt the bucket and kick the bucket to be weakerthan between the two phrases mentioned previously. Gibbs et al. concludes thatthe relative lack of meaning disruption is due to the fact that some synonymspreserve the metaphorical relation between an idiom's individual component andits real-world referent. But their experiments did not involve speaker's com-prehension of variant idioms, and thus they cannot be regarded as evidence thatstrings like burst the question, pop the request, and burst the request are indeedall interpreted as propose marriage. Therefore, the results do not show that theparticular sense of the (polysemous) word pop, which is related to ask, also isrepresented in the lexicon by the word burst.

Glucksberg, McClone, and Cacciari (1991) cite a few real-life utterances such

12. THE DETERMINER IN ENGLISH IDIOMS

as Convicted minimalist spills bean and eat the golden goose. They tested speak-er's comprehension of altered idioms, noting that although the canonical forms ofidioms are processed fastest, the comprehension of altered idioms takes no moretime than the comprehension of literal strings. Glucksberg et al. conclude thatvariant idioms in fact are processed in the same way as nonfigurative speech,because the constituents of idioms, through frequent use, often have come to beassociated with their appropriate meanings, for which ordinary synonyms thencan be substituted via standard linguistic processing.

Although this hypothesis seems quite plausible and accounts for the process-ing times that Glucksberg et al. ( 1991) measured, it does not address the questionas to why a literal word often can be substituted for an idiom constituent whereasthe idiom constituent is not necessarily free to be used outside the idiomaticphrase. Thus, although speakers may understand the phrase He spilled the se-cret(s), it is not clear that they will understand a phrase like He told me somebeans about her. In other words, the close association of the idiomatic and theliteral word with one given concept does not necessarily permit a two-wayexchange of these two words expressing that concept. This indicates that there isa difference between denoting nouns in idioms and conventional, free metaphors.A free metaphor such as angel, meaning roughly kind person, can occur outsideof specific idiomatic phrases; it has the same distribution as the correspondingliteral NP kind person. But nouns like beans and ice cannot be substituted in allcontexts for secret(s) and tension/shyness. The same can be said for idiomaticuses of verbs, which often receive the intended meaning only in the presence of aspecific NP. Thus, the sense of drop corresponding to write is available only inthe context of a line but not in sentences like She dropped her grandparentslherterm paperlbeautiful French/legiblylall afternoon.

Nevertheless, the results of Gibbs et al. (1989) and Glucksberg et al. (1991)show that the canonical forms of idioms are subject to considerable mor-phological and lexical variation, and the question arises as to how such variationsare both generated and constrained.

Our knowledge of words and idioms, that is, our lexical knowledge, is part ofour language competence. But much of our linguistic behavior must be explainedin terms of performance, that is, the way language is used in actual communica-tion (this distinction is due to Chomsky, 1956, and was also drawn by Saussure,who contrasted langue and parole). Competence guides our intuitions and judg-ments about distinguishing between grammatical, well-formed utterances, andungrammatical or ill-formed ones. By contrast, our language performance ac-counts for both the production and the successful processing of grammaticalerrors, unfinished sentences, pronunciation mistakes, and puns.

Idioms are subject to word plays and puns that result in alterations of theircanonical forms just as much as other strings. Idiosyncratic variations are createdby speakers to fit certain discourse-specific contexts. A priori there is no limit tothe changes one could make to an idiom and still expect one's interlocutors to

277

278 FELLBAUM

process it. Given the appropriate context, it does not seem unreasonableto assume that even an utterance like Uander great pressure, he finally disgorgedthe legumes can be understood by a listener who is aware of the speaker'shumorous intent.

An analysis of the lexical, morphological, and syntactic flexibility of idioms,and the possible systematic constraints on this flexibility would represent a veryconsiderable undertaking. In the absence of such a study, this chapter limits itselfto a discussion of the canonical forms of idioms and the alterations that arepossible in the absence of specific contexts. In the literature on idioms, authorscommonly make judgments about which changes preserve the figurative mean-ing of idioms and which ones do not, and readers generally accept these judg-ments. For example, people do seem to agree that have a certain axe to grind is apermissible alteration of the idiom have in axe to grind, and they also agree thathave the axe to grind does not preserve the intended idiomatic sense. It isimportant to note that such judgments are made in the absence of any context onthe basis of our language competence only. These judgments, and speakers'agreements about them, clearly set off the competence-ruled data that are thesubject of this study from the potentially open class of possible variations,including those studied by Gibbs et al. ( 1989) and Glucksberg et al. (1991). Wetake the fact that speakers uniformly agree about the canonical forms of idiomsand the limits on their context-free variations as a license to limit our study and todisregard the idioms' many possible discourse-conditioned alternations.

The variations in VP idioms that are examined here are limited to the deter-miner. These determiner changes are often acceptable in context-free uses of theidiom, much like subject-verb agreement or changes in the tense of the verb.Other variants of the determiner, although not created for humorous purposes,are discourse-conditioned and show the speaker's analysis of the NPs as denotingentities. We see that the determiners of NPs in many idioms can be analyzedalong the rules of literal language, and that the nature of the determiner fre-quently reveals the metaphoric status of the nouns.

THE REFERENTIAL STATUS OF NPsIN IDIOMS

Unlike denoting nouns like beans in the expression spill the beans, nouns likedust and bucket in bite the dust and kick the bucket are not mapped onto ameaning in the lexicon and are therefore semantically empty. Their literal in-terpretations (i.e., pail and fine dirt particles, respectively) are not availablehere, nor can the nouns as separate parts of the idioms receive a metaphoricinterpretation (such as death) that is related to the meaning of the idiom as asemantic unit. Each of these noncompositional VP idioms as a whole is linked to


the meaning of the intransitive verb die. We call NPs like bucket and dust in theseidioms "nondenoting."

The distinction between denoting and nondenoting NPs in figurative speech isfrequently made on a purely intuitive basis. Although it does seem intuitivelyobvious that dust in bite the dust receives no independent figurative meaningoutside the VP, and although it seems similarly intuitively obvious that hatchet inbury the hatchet has some independent figurative meaning, these intuitions arehard to back up, because the use of the noun with the particular figurative sense isusually limited to a single specific occurrence.2

Another kind of evidence for the denoting or vacuous nature of an idiom NPoften has come from the syntactic shape of the idiom: When the idiom consists ofa verb plus a direct object, and its literal equivalent is an intransitive verb, thenthe fact that the idioms NP cannot be matched onto a literal NP is thought toindicate the nonreferring status of the former: There is simply no way to "trans-late" the noun. However, the nondenoting status of a NP is not linked necessarilyto the lack of a NP in the literal equivalent of the idiom. Although die isundoubtedly the most felicitous, and most frequent, word for expressing thesense of kick the bucket or bite the dust, the same meaning also could beexpressed by cease all biological activity (Collins dictionary). This VP struc-turally matches the idiomatic VPs, yet we do not interpret bucket or dust to standfor biological activity here. (This raises the thorny issue of how to determine theliteral equivalent of an idiom, especially in less clear-cut cases than these.)3

Consider also the expression eat humble pie, which is glossed in Boatner et al.(1975) as accept insult or shame or admit your error. Here, too, the idiom and itsliteral counterpart are structurally (syntactically) equivalent. Yet, humble pieseems to have no metaphorical relation to insult, shame, or error. Conversely,the noun in hit the road appears to be semantically interpretable (literally, infact), but it does not seem to fit into the literal equivalent of this idiom (leave,depart).

The strongest support for the semantic status of idiom NPs has come from thesyntactic behavior of idioms. Idioms whose nouns are presumed to be seman-tically empty, such as kick the bucket and bite the dust, turn out to be highlyfrozen syntactically; this is commonly taken as the reflex of the fact that the NPs

2There are very few cases of idiom constituents that preserve (part of) their figurative meaning inmore than one context. An example might be beans, meaning information, in both spill the beans andI don't know beans. By contrast, sack has a completely different meaning in each of the idioms hit thesack, get the sack, and leave someone holding the sack. (But cf. Ruhl, 1989, who argues formonosemy across literal and figurative language.) Wasow et al. (1983) claimed that strings in pullstrings is semantically sufficiently transparent to be used independently of pull; however, they wereunclear on the constraints of such uses.

3 Nunberg (1978) notes that give up the ghost can passivize, which he interpret as evidence thatspeakers assign an interpretation to ghost.

280 FELLBAUM

here have no referent. By contrast, idioms with denoting nouns tend to beavailable for syntactic movement operations such as passive, and tolerate theinsertion of lexical material like adjectives. The flexibility of an NP in an idiomappears indeed to be tied to the referential status of the NP (Newmeyer, 1974;Nunberg, 1978; Wasow et al., 1983; and others), although this correlation is notalways straightforward. 4 We show that, similarly, the nature of the determiner inthe dictionary form of the idiom and the determiner's alterability yield importantclues to the referential status of the NP, as well as to its specificity, and itsknown, presupposed or inferred status.

IDIOMS WITH A DEFINITE DETERMINER

In literal language, the definite determiner is polysemous; that is, it can precedenouns with several different kinds of semantic status. With a denoting NP, theprincipal uses of the definite determiner in the literal language can be summed upas follows: (a) unique: The sun rises in the east; (b) institutional: I am going tothe post officelThe radio broadcast the news of the demonstration; (c) inferablefrom general knowledge, or pragmatically presupposed: Ring the bell twice whenyou come for dinnerlAttach this antenna to the television; (d) inferable from thepreceding discourse: John told a joke. The punchline wasn't really funny; (e)known from previous mention in the preceding discourse: We went to see aplay. . . . The play was terribly boring; (f) definite, specific: The biology text-book we used last year is bad. The NPs in their definite, unique, inferable, andknown uses all denote specific referents; "institutional" NPs have nonspecificreferents. Most of these uses of the definite determiner also can be distinguishedin VP idioms.5

The nouns referring to unique entities in the idioms that follow are not meta-phors, but receive the same literal interpretation that they would be assignedoutside the idioms. Just like in nonfigurative contexts, they are preceded by adefinite determiner.

16. He promised her the moon.17. The bank robber told the teller to reach for the sky.

4 Wasow et al. (1983) observe that whether or not idiom NPs can pronominalize indicates theirmetaphorical status: "We thought tabs were being kept on us, but they weren't." But note that someapparently denoting NPs do not pronominalize so, asily: "John and Sam finally buried the hatchet,and then Bob and Jane buried it." This indicates that pronominalization is not a sufficiently reliableindicator of a noun's status as a metaphor.

50ther nonspecific referents preceded by a definite article are generic NPs as in The astrolabe isone of mankind's great inventions and specific indefinites like The person who guesses the rightnumber will be an instant millionaire. There is no evidence for such generic or specific indefinitereferents in VP idioms.


The definite article here cannot be exchanged for an indefinite one, a negative, ora quantifier; this confirms the literal interpretation of these nouns:

18. He promised her {a/no/every} moon.19. The bank robber told the teller to reach for {a/no/each} sky.

These nouns are like proper names and refer to specific individual entities. Thereason that unique entities are referred to literally in idioms may be the follow-ing: The literal equivalent of a noun in most idioms is somewhat vague andadmits of several semantically similar interpretations. Thus, axe in have an axeto grind may refer to a grievance or an ulterior motive; the exact meaning is notimportant, because the meaning of the idiom as a whole is not dependent on sucha fairly subtle difference. However, the meaning of an idiom like that in Sentence16 depends crucially on the uniqueness of the noun's referents, and with uniquereferents, such vagueness of interpretation is not possible. A nonliteral noun witha unique referent would be difficult to interpret, because the word-meaningrelation in an idiom tends to be valid only for the one particular context. But notethat conventional metaphors, whose exact meaning is less vague than that ofidiom NPs, can occur in idioms as well: Heaven(s) commonly stands for sky inmany contexts, and also occurs in the idiom go to heaven; the (great) maker, aconventional metaphor for God, also has this meaning in the idiom meet one'smaker.

Note the use of the indefinite article in accordance with the rules of literallanguage in He only goes fishing once in a blue moon: Blue moons, a naturalphenomenon, are rare, but not unique.

The use of the definite article in such nonspecific NPs as the post office, thesupermarket, and the TV could be called institutional (somewhat differently fromthe use of that term in Quirk, Greenbaum, and Svartvik, 1985). It also can befound with NPs in some VP idioms, which receive this institutional interpreta-tion:

20. This court ruling will {turn/set/put} the clock back.21. In this house, the wife wears the {pants/trousers}.22. John's wife is the one who brings home the bacon.23. After dinner, he decided to hit the road.

The clock in Sentence 20 refers to progress; the pants/trousers in Sentence 21stands for (traditionally male) authority or power, and the bacon in Sentence 22means financial support. These senses are institutional in that their referents areunderstood and known without having been mentioned previously, similar to thepost office in a (nonidiomatic) sentence like I went to the post office this morning.

282 FELLBAUM

The road in Sentence 23 is not a metaphor, but is interpreted literally in a waysimilar to post office.

When the definite article is exchanged for another kind of determiner, thereferent of the NP no longer receives the institutional interpretation. This is thecase for both literal and figurative NPs:

24. I went to {a/each} post office this morning.

25. This ruling will set (all) clocks back.26. In this house, the wife wears 'trousers/pants}.27. John's wife brings home some bacon.28. After dinner he decided to hit that road behind our yard.

A somewhat similar case is presented by the pragmatically presupposed referentsin the following idioms:

29. When John heard the news. he hit the {roof/ceiling}.30. Mary went through the {roof/ceiling} when her rent was raised.31. He pulled the plug on her activities.

The definite determiners here are justified because of the pragmatically presup-posed, and thus "given," referents of the NPs, which resemble those in the(nonfigurative) strings listed next:

32. When you get home, put this in the freezer.33. This antenna must be attached to the TV.

People commonly assume the existence of a freezer, a roof, a ceiling, plugs, anda TV in everybody's life. These quasi-generic, nonspecific referents therefore arepresupposed and constitute a kind of world knowledge. They are referred towithout having occurred prior in the given discourse. The substitution of anotherkind of determiner in both literal and figurative speech removes this reading.6

Compare the following sentences:

34. When you get home, put this in {a/each/some} freezer.35. When John heard the news, he hit a {roof/ceiling}.

Whereas such referents as roof, ceiling, and freezer can be interpreted indepen-dently of a given discourse situation or context, the definite determiner in some

6 The common use of the "generic" possessive your illustrates the assumption or presuppositionon the part of the speaker that his interlocutor possesses certain things: Put this in yourfreezer. IThisantenna must be attached to your TV.


idioms precedes a NP whose referent is inferable from the context or discourse inwhich it occurs:

36. At every party John manages to break the ice with his jokes.

The existence of ice, that is, tension and uneasiness between people in a socialsituation, is inferable from party.7 The use of the definite article parallels that ina similar sentence like Sentence 37 that follows next, which receives a literalinterpretation; just as party licenses the interpretation of the ice, lecture makesthe notes inferable:

37. When John gives a lecture, he always forgets the notes.

An indefinite article in both kinds of sentences would result in an interpretationof the NP that is disjoint semantically from the preceding discourse. The in-terpretation of such indefinite NPs accounts for the oddness of Sentences 38 and,especially, 39:

38. At every party, John manages to break ice with his jokes.39. When John gives a lecture, he always forgets notes.

The string break the ice receives an idiomatic reading naturally in a contextwhere such tension is inferable; Sentence 40 that follows is odd, even thoughsocial uneasiness is not inconceivable even in a grocery store:

40. At every grocery store, John manages to break the ice.

Contrast this with the distribution of a noncompositional idiom like kick thebucket, which is not context-dependent and which can occur in virtually anycontext, because the bucket does not refer to an inferable entity:

41. John kicked the bucket at the {grocery store/birthday party}.

The inferences we make, such as interpreting ice on the context of party, are aconsequence of the structure of our mental lexicon. There is strong evidence thatsemantically related words and concepts are stored together (Gross, Fischer, &Miller, 1989; Miller, 1969; and others). And in natural, coherent communica-tion, related words tended to co-occur, (Charles & Miller, 1989; Fellbaum, 1992;Justeson & Katz, 1991) because we usually communicate about a given topic or a

7Evidence that ice stands for tension, uneasiness also comes from the fact that it is a mass noun,just like its literal counterpart, and cannot pluralize: John breaks the ices at every party is no longerdistinctly odd.

284 FELLBAUM

range of related topics. Whereas the nonidiomatic word form ice with its associ-ated meaning,frozen water, is semantically connected to, and occurs naturally incontext with, such words as cold, winter, and so on, the same word ice that isassociated with the meaning tension, uneasiness, is connected to words denotingsuch concepts as party, strangers, shyness, social interaction, and so forth.8

It follows that the entities that can be inferred from the context of a figurativenoun are more likely to be semantically linked to that figurative meaning, ratherthan to the literal one. Thus, sequences like Sentences 42 and 43 receive either aliteral reading or can be interpreted as humorous variants of the underlyingidioms. In fact, the continuing discourses must be controlled carefully to permitsuch humorous readings; not every discourse that is compatible with the literalinterpretation is possible with the humorous idiom variant: It is not clear what thepieces in Sentences 42 and the fat in Sentence 43 would refer to if the ice wereinterpreted figuratively as components of tension, shyness and earnings, respec-tively:

42. He broke the ice at the party. There were many small pieces left.43. John's wife brings home the bacon. She always removes the fat.

Note also Pulman's (1986) discussion of the impossibility of substituting a spe-cific hyponym for the noun in an idiom, as in the sentence *John put the tabbyamong the pigeons.

NPs that are definite because they were previously mentioned are common inliteral language (Clark & Haviland, 1977; Prince, 1981; and others), but not inidiomatic language. This is necessarily so because the figurative NP receives itsinterpretation only by virtue of its setting within the idiomatic VP; the NP, notbeing a conventional metaphor, cannot be repeated with its intended meaning inanother context. Thus, whereas Sentence 44 is a plausible discourse, Sentence 45is not:

44. We saw a play the other night. . . . The play was boring.

45. John has a bone to pick with lBob. . . . The bone is small.

We can distinguish yet another kind of idiom NP with a definite determinerthat corresponds to a semantic class of NPs in literal language. These are NPswith a specific referent known to the discourse participants. The fact that the NPis interpreted figuratively makes no difference; this use of the article is the sameas it would be with an NP that receives a literal reading, that is, with any

S1n Glucksberg et al's (1991) model of idionr. representation, both literal and idiomatic senses areavailable and permit speakers to create and interpret variants based on literal properties of thecomponents. But the figurative and the literal NP, that denote a given concept are not equal in thatthey are not interchangeable in all contexts.


synonym of the known referent. For example, the beans in the VP spill the beansmust refer to a secret whose existence and contents are known to the discourseparticipants. Thus, Sentence 46 is an acceptable sequence, whereas Sentence 47,where the contents of the secret is made known to the previously ignorantlisteners, is odd:

46. John spilled the beans (about his girlfriend).47. John spilled the beans that his girlfriend once worked for the CIA.

The noun secret can be both definite or indefinite; if it is definite, it obeys thesame constraint as the beans:

48. John told {a/the} secret (about his girlfriend).49. John told the secret that his girlfriend once worked for the CIA.

Thus, the beans in the idiom refers to a particular, known secret or confidentialinformation. Note that because this noun refers to a specific secret it can bemodified only by a quantifier that is semantically compatible:

50. ?John spilled {some/many/hundreds of} beans about Mary.51. John spilled all the beans about Mary.

Note also that due to the specificity of the beans, only the VP-external negation ispossible; the VP-internal negation in Sentence 53 is decidedly odd:

52. John didn't spill the beans about his girlfriend.53. John spilled no beans about his girlfriend.

The idiom bury the hatchet can be analyzed in the same way: Sentence 54 can beuttered felicitously only when the discourse participants are aware of the exis-tence and nature of the hostile relationship between John and Bob:

54. Yesterday, John and Bob buried the hatchet.

By contrast, Sentence 55 sounds odd, because the idiom is used without thereferent of the NP having been established in the listeners' minds:

55. Yesterday, John and Bob buried the hatchet about property lines.

In the cases of idioms like spill the beans and bury the hatchet, we saw thatthe definite article is justified by the semantic nature of the noun's referent, andthat the determiner choice follows the same rules as in literal language. In every

286 FELLBAUM

case, the definite article is fixed and cannot be exchanged for an indefinite or anegative determiner, which would remove the presupposition of there existing asecret or a relation of enmity:

56. Bob and John buried {a/no} hatchet.57. John spilled (no) beans about his girlfriend.

The NPs in these idioms often can be modified by a demonstrative adjectiveso long as the referents have been activated in the given discourse. Because theuse of the demonstrative adjective is discourse-dependent, demonstratives do notoccur in the dictionary forms of idioms. Note the following examples:

58. Bob and John finally buried that old hatchet.59. We'll cross that bridge when we get to it.60. I'll take THAT bull by the horn.61. Those windmills not even he would tilt at. (Wasow et al.'s Sentence 13)

The demonstrative here always modifies a "given" NP; it seems that its deicticmeaning is not available for NPs in idioms. The reason may be that the referentsof NPs in idioms generally denote abstract referents lacking a physical extension,rather than concrete entities that can be pointed to.

The particular word used to denote the referent may be new in the discourse ifthe idiom is used for the first time, but the associated concepts enmity, obsta-cleldifficulty, and imaginary opponent must have been activated in the discourse;the nouns in the idioms function just like synonyms for other words expressingthe same concepts. The use of the demonstrative here, then, is no different fromthat in literal language.

The remaining cases of definite determiners in VP idioms do not fall into anyof the uses that have been distinguished for literal language. The nouns here donot have generic, unique, specific, known, or inferable referents and seem toviolate the normal rules of determiner selection. Examples are kick the bucket,bite the dust, chew the fat, shoot the breeze, give the slip to, and many others.The nouns in these idioms cannot be interpreted either literally or metaphorically,and appear to be nondenoting. This fairly large class of idioms has been noted forboth its semantic opacity and its syntactic frozenness with respect to passiviza-tion, insertion of lexical material, and so on. (Nunberg, 1978; and others). Gibbset al. (1989) also show that these idioms do not tolerate lexical substitutionwithout losing their figurative meaning. The emptiness of the noun that has beendemonstrated by the syntactic inflexibility of the idiom is confirmed by the factthat the definite determiner is not accountable for by the usual principles. We callthe use of the definite determiner here the idiomatic use, as it seems to be the oneuse of the definite article that has no correspondence in literal language.


The syntactic and lexical rigidity of these idioms that has been noted (Cutler,1982; Fraser, 1970; Gibbs et al., 1989; Newmeyer, 1974; and others) also isfound within the NP, where the definite determiner cannot be altered without aresultant loss of the figurative meaning:

62. Mary finally gave the slip to her boyfriend.

63. Mary finally gave that particular slip to her boyfriend.64. Mary finally gave a certain slip to her boyfriend.

65. Mary finally gave {no/some/every} slip to her boyfriend.

We suggest that the idiomatic use of the definite determiner has a specificfunction, namely, to alert the hearer to the nonliteral nature of the expression.The hearer cannot assign any of the standard meanings of the definite article tothe NPs in such idioms, and, by default, interprets it as idiomatic. Note that if thearticle preceding such nondenoting nouns were indefinite, the listener wouldinterpret them according to the rules of literal language, that is, as entities newlyintroduced into the discourse for which a referent must be established, as in Johnkicked a bucket. We see that, in fact, there exist very few VP idioms where anondenoting noun is preceded by an indefinite article.9

VP IDIOMS WITH INDEFINITE NPs

Dictionaries commonly classify VPs such as take a nap and have a drink asidioms. These constructions are characterized by the presence of a semanticallyempty "light" verb such as do, have, make, or take, and a singular indefinite NP;they denote actions that also can be referred to by verbs that are homophonouswith the nouns in the VPs (to nap and to drink, in the examples just mentioned).Because these VP constructions are productive in English (but see Wierzbicka,1982, for a principled account of the constraints on their productivity), andbecause of their semantic transparency, we do not consider them as true idioms.These VPs also exhibit a high degree of semantic and syntactic flexibility, similarto nonidiomatic strings. They could be classified as the kinds of pseudoidiomatic

9This hypothesis could be checked in an experiment monitoring the on-line processing of idiomslike bite the bullet and chew the fat to determine whether hearers assign a nonliteral reading to theexpression soon after they hear the definite article. Cacciari and Tabossi (1988) and Tabossi andCacciari (1988) raise the question of determining the key in an idiom, that is, the particular constitu-ent that alerts the hearer to the nonliteral meaning of an expression; in these idioms, the key may benot a content word, but the definite determiner. Another test might consist in presenting made-upidioms (or idioms literally translated from another language) with definite NPs to speakers and letthem judge whether or not the string is to be interpreted literally, figuratively, or whether it is anonsensical string. We would predict that even though the meaning of the made-up idiom is opaque tothe hearers, they should be able to distinguish it as an idiom, rather than as a nonsensical string.

287

288 FELLBAUM

constructions (like the X-er the Y-er) studied by Fillmore, Kay, and O'Connor(1988).

Apart from these constructions, there are relatively few VP idioms whosedictionary form shows an indefinite Nil. This makes sense given that the listeneris likely to interpret an indefinite noun literally, assigning it a "new," nonspecificreferent, rather than interpreting it idiomatically. There is no a priori reason whythe listener would not follow the normal rules for interpreting an indefinitearticle, which precedes a noun that cannot be linked to an entity that is old in thediscourse or otherwise presupposed. The listener is likely to assign a referent tothe noun on the basis of its literal interpretation, unless both the larger contextand the verb in the idiomatic VP are semantically sufficiently incongruous toindicate that a literal interpretation is not appropriate here.'I

Even when the idiom is interpreted figuratively, the indefinite article main-tains its function, and the referent of the NP is is assigned a new reading. This isthe case, for example, in the sentence that follows:

66. Mike smelled a rat.

Consider also the following idioms:

67. Mike danced up a storm.

68. You are building castles in the air.69. When she heard the news, she made a scene.

Here, too, the indefinite article is used in accordance with the rules of literarylanguage, signaling that the noun is to be interpreted as new. The verbs dance up,build and make denote the "creation"' of something new. Similarly, the indefinitearticle in idioms like turn over a new leaf and sing a different tune is used justlike in literal language, with the adjectives new and different reenforcing thesemantics of the NP.

The entities denoted by the objects of creation verbs do not exist prior to theaction denoted by the verb. Therefore, they cannot be referred to by specificNPs, and do not tolerate the insertion of adjectives like specific or certain:

70. She was singing a specific new tune.71. Mike danced up a certain storm.72. She turned over a specific new leaf.

Sentence 74 shows that the same holds true for literal language. In Sentence 73,the verb is not one of creation, so the object noun may be specific. By contrast,

10The generic use of the indefinite article, as in 4 lion won't attack when it has justfed, does notoccur in idioms.


in Sentence 75 the passive topicalizes the noun, which results in its referenthaving an (indefinite) specific reading:

73. Mary bought a certain rug.

74. Mary wove a (*certain) rug.75. A certain rug was {woven/bought} by Mary.

Similarly, negation in these idioms must have scope over the entire VP. In literallanguage, negation of only the argument often implies a contrast, but suggeststhat an argument is presupposed; negation over the entire VP lacks such apresupposition, as Sentences 76-79 show (see Babby, 1980, for a discussion ofcontrastive negation). Although Sentence 80 may mean that Mary is weavingsomething other than a rug, Sentence 81 does not seem to imply that Mikedanced up something other than a storm. There does not seem to be any presup-position in Sentences 81-83-that Mike danced up something, Sue made some-thing, and Larry will turn over something-that is responsible for the non-idiomatic reading. The reason that these sentences are not meaningful is that theverbs do not receive an interpretation independent of their noun arguments:

76. Mary isn't weaving a rug.

77. Mike didn't dance up a storm after all.78. Fortunately Sue did not make a scene.79. Larry won't turn over a new leaf.

80. Mary is weaving no rug.81. Mike danced up no storm after all.82. Fortunately Sue made no scene.83. Larry will turn over no new leaf.

It might be argued that the nouns in these idioms are semantically empty, whichwould account for their inflexibility. However, note that some of these theseidioms can undergo passivization or topicalization, which has been taken toindicate that people assign an interpretation to the noun (Nunberg, 1978; andothers):

84. A new leaf was turned over when the management changed.

85. What a scene she made when she heard the news!

The fact that the use of the indefinite article here conforms to that of literallanguage indicates that the NPs in these idioms function as metaphors and have areferent, but only in composition with the particular verbs. The relation between

290 FELLBAUM

the figurative NP and its referent here is maintained by the use of the indefinitearticle.

Consider next the idiom in the following sentence:

86. John has a {bone/crow} to pick with Bob.

The NPs here have a referent whose meaning is rather close to the hatchet in burythe hatchet; namely, a point of contention or quarrel. But unlike in that idiom,the referents of bone and crow are new to the listener, as the indefinite articleindicates. Indeed, the sequence in Sentence 87 is fine, and should be comparedwith the less felicitous Sentence 88:

87. John has a {bone/crow} to pick with Bob about property lines.

88. John and Bob buried the hatchet about property lines.

The same analysis applies to the idiom have an axe to grind.Although the referents of the NPs in these idioms are indefinite, that is, not

known to the listener, they are specific. This can be seen by the fact that theidioms tolerate the insertion of adjectives that induce a specificity reading:

89. John has a {specific/certain/particular} bone to pick with Bob.90. Mary had a {specific/certain/particular} axe to grind with her boss.

The NPs in these idioms can be negated and quantified:

91. John has {several/no} bones to pick with Bob.92. Mary had {more than one/no} axe to grind with her boss.

In contrast to the idioms in Sentences 76-79, the negation and specificity dataindicate that the interpretation of the nouns' referents here do not seem to dependclosely on the verbs.

A small number of idioms have indefinite NPs without a referent:

93. It seems John has a chip on his shoulder.94. We had a ball watching the comedians.

The NPs here cannot be negated or modified either by quantifiers or by adjectivesthat denote specificity:

95. We had no ball last night.

96. It seems John has several chips on his shoulder.97. John seems to have a specific chip on his shoulder.


The question arises why these idioms do not have a definite determiner thatprevents the hearer from assigning a new referent to the NP, such as in the idiomskick the bucket and bite the dust. To begin with, a survey of an idiom dictionarysuch as Boatner et al. (1975) shows up far more opaque idioms with definite NPsthan with indefinite NPs, which make up a very small class. Second, althoughthey are completely inflexible, such nouns as ball and chip seem semanticallyslightly more transparent than do bucket and dust. Ball could be glossed withgood time (which, in the same construction, can be focused in passivization andthus apparently has a referent); chip seems to refer to grudge. Although there isno evidence that people analyze the nouns in this way, such an analysis is leastplausible here, unlike in idioms of the kick the bucket type.

Some VP idioms have bare plurals, for example, cut corners and keep tabs on.These are plural indefinite, and like singular indefinites, they tend to be flexibleif the noun has a referent. Thus, the corners in cut corners refer to shortcuts, andthey can be negated or quantified:

98. He cuts {lots of/no} corners.

A much-discussed idiom VP is pull strings/wires; the relation between the sur-face NP and its meaning, connections, is particularly salient here. As has beenobserved (Wasow et al., 1983; and others), this idiom shows great syntacticflexibility, and the NP can be modified in any way compatible with its semantics:

99. Her mother pulled {many/no/these/certain} strings.

By contrast, tabs refers to watch, guard, and is not quantifiable:

100. *The FB kept many tabs on him.

Thus we see that the behavior of the indefinite determiners in idioms NPs paral-lels that in literal language.

BARE NOUNS

Many VP idioms contain a singular count noun with zero determiner, and, by thecriteria of literal language, thus could be called "asyntactic" (Cruse, 1986, andothers):

101. In this war, the generals {saved/lost} face.102. They could not make head or tail of his story.103. John {took/lost} heart when he heard the news.

291

292 FELLBAUM

One could argue that the nouns have become incorporated into the verbs (Baker,1988), that is, that they form a syntactic unit with the verbs. The idioms aresyntactically frozen and cannot undergo such movement operations as passiviza-tion. However, the NPs here seem to be metaphors that refer to a mass noun:Face is synonymous with (personal) honor; head or tail stands for sense; andheart has the sense of courage. Given these alternate, metaphoric senses of thenouns, idioms such as in Sentences 101-103 are no longer asyntactic; a singularmass noun standardly occurs with zero determiner. Thus, the asyntacticity arisesonly under the literal interpretation of the noun, when it is a count noun.

The nouns in such idioms can be preceded by determiners that are compatiblewith their semantics. For example, a quantifier can be inserted when the referentof the NP is quantifiable:

104. They couldn't make much head or tail of his story.

105. The teacher took ({little/no}) stock in the student's report.

106. The movement lost ({much/no}) ground in the '90's.107. He had to kiss ({quite a bit of/his colleagues'}) ass to get the promo-

tion. " I

IIThe idiom kiss ass does not follow the usual pattern of idioms with body-part nouns: Ex-pressions like stab someone in the back, pat someone on the back, and hit someone in theface displaywhat is known as possessor ascension syntax, where the "possessor" NP is the direct argument of theverb and the "possessed" body part appears in a prepositional phrase (Fox, 1981; and others). In theliteral language, but not in the idioms, an alternative syntax is possible:

la. John kissed his daughter on the forehead.lb. John kissed his daughter's forehead.2a. John stabbed his best friend in the back.2b. John stabbed his best friend's back.3a. The news hit her in the face.3b. The news hit her face.

It has been noted (Fox, 1981) that the possessor ast ension syntax seems to express a greater degree ofaffectedness of the possessor. In body-part idioms,, the possessor always is affected by the action,whereas the body-part noun usually does not receive a literal (and affected) interpretation: In sentence2a, under the idiomatic reading, John's bestfriend is being acted upon, but his back is not. This mayexplain the obligatory possessor ascension syntax in the idioms; it would be evidence for Nunberg's(1978) claim that their syntax may be sensitive to fairly subtle semantic properties of the idioms. Butkiss ass, when it occurs with a possessor, behaves differently; only the configuration with ass as thedirect argument is idiomatic, whereas the possessor ascension version must be interpreted literally:

4a. John kissed his colleagues' ass.4b. John kissed his colleagues on the ass.

Note also that the possessor does not have to have a surface form in this idiom, as sentence 107shows. Unlike the body-part NPs in such idioms as stab someone in the back and hit someone in theface, ass in kiss ass is interpreted as a mass noun and therefore does not follow the canonical syntaxof body-part language.


The nouns with zero determiner in the following idioms have no referent:

108. His opponent had to eat {humble pie/crow}.

109. The salesman started to talk turkey.

These highly frozen idioms resemble the class of idioms with definite deter-miners that includes bite the dust and shoot the breeze, and those with indefinitedeterminers like have a ball. Predictably, they tolerate no NP-internal modifica-tion whatsoever:

110. His opponent had to eat {some/no/a piece of} humble pie.111. The salesman started to talk {a lot of/no} turkey.

CONCLUSION

An examination of the nouns in many VP idioms shows that their semantics arereflected in the determiners they are compatible with. The range of determiners isconstrained by the same semantic and pragmatic principles as in literal language.

The definite determiner occurring in the dictionary forms of many idioms isinflexible when it modifies nouns that can be shown to have unique, institutional,or discourse-determined referents. When these referents have been activated inthe discourse, demonstratives may be substituted-a pragmatic principle oper-ative in literal language.

The indefinite article in the VP idioms we investigated also can be shown tofollow from the nouns' semantics; it is used with new or specific indefinitereferents. As predictable from the corresponding case in literal language, quan-tifiers or negative adjectives can be substituted.

The zero determiner occurs in apparently asyntactic idioms with count nouns;however, these nouns have figurative meanings corresponding to mass nouns,which are compatible with zero determiner. The interpretation of these nouns asmass nouns is further supported by their compatibility with certain quantifiers.

In sum, the analysis of the determiners in the kinds of VP idioms we examinedsupports a compositional of idioms, and, in particular, serves to test and strength-en intuitions about the referential status and the meaning of metaphoric nouns.

The class of idioms where these principles fail shows inflexibility not onlywith respect to the determiner but also with respect to lexical changes andsyntactic operations such as passivization, indicating that the noun has no seman-tic content. We suggested a special, pragmatically based function for the definitedeterminer that is found in most of these noncompositional idioms.

293

294 FELLBAUM

ACKNOWLEDGMENTS

Preparation of this chapter was supported in part by contract N00014-86-K-0492with the Office of Naval Research (ONR) and in part by a grant from the James S.McDonnell Foundation. The views and conclusions contained herein are those ofthe author and should not be represented as official policies of ONR, the McDon-nell Foundation, or Princeton University. We thank Cristina Cacciari and PatriziaTabossi for valuable comments, and members of the Cognitive Science Laborato-ry for help with some of the data.

REFERENCES

Babby, L. (1980). Existential sentences and negation in Russia..Ann Arbor: Karoma.Baker, M. C. (1988). Incorporation: A theory qf grammatical function changing. Chicago: Univer-

sity of Chicago Press.Boatner, M., Gates, J., & Makkai, A. (1975). A dictionary of American idioms. New York: Bar-

ron's Educational Series.Bobrow, S. A., & Bell, S. M. (1973). On catching on to idiomatic expressoins. Memory & Cogni-

tion, 1, 343-346.Cacciari, C., & Tabossi, P. (1988). The comprehension of idioms. Journal of Memory and Lan-

guage, 27, 668-683.Chomsky, N. (1956). Syntactic structure. The Hlague, Netherlands: Mouton.Clark, H., & Haviland, S. (1977). Comprehension and the given-new contrast. In R. Freedle (Ed.),

Discourse production and comprehension (pp. 1-40). Norwood, NJ: Ablex.Cruse, D. A. (1986). Lexical semantics. Cambridge, England: Cambridge University Press.Cutler, A. (1982). Idioms: The colder the older Linguistic Inquiry, 13, 317-320.Fillmore, C. J., Kay, P., & O'Connor, M. C. (1988). Regularity and idiomaticity in grammatical

constructions. Language, 64, 501-538.Fellbaum, C. (1992). Co-occurrence and antonymy (Technical Report No. 52). Princeton, NJ:

Princeton University, Cognitive Science Laboratory.Fox, B. (1981). Body part syntax: Towards a universal characterization. Studies in Language,

5,323-340.Fraser, B. (1970). Idioms within a transformational grammar. Foundations of Language, 6, 22-42.Gibbs, R. W., & Nayak, N. P. (1989). Psycholinguistic studies on the syntactic behavior of idioms.

Cognitive Psychology, 21, 100- 138.Gibbs, R. W., Nayak, N. P., Bolton, J. L., & Keppel, M. E. (1989). Speakers' assumptions about

the lexical flexibility of idioms. Memory & ignitionn, 17, 58-68.Glucksberg, S., McGlone, M., & Cacciari, C. (1991). Semantic productivity and idiom com-

prehension. Manuscript submitted for publication.Gross, D., Fischer, U., & Miller, G. A. (1989). The organization of adjectival meanings. Journal of

Language and Memory, 28, 92-106.Justeson, J. S., & Katz, S. M. (1989). Co-occurrences of antonymous adjectives and their contexts.

Computational Linguistics, 17, 1 -19.Miller, G. A. (1969). The organization of lexical memory: Are word associations sufficient? In

G. A. Talland & N. C. Waugh (Eds.), The pathology of memory (pp. 223-236). New York: Aca-demic.

Charles, W. G., & Miller, G. A. (1989). Contexts of antonymous adjectives. Applied Psycho-linguistics, 10, 357-375.


Newmeyer, F. L. (1974). The regularity of idiom behavior. Lingua, 34, 327-342.Nunberg, G. D. (1978). The pragmatics of reference. Bloomington: Indiana University Linguistics

Club.Prince, E. (1981). Towards a taxonomy of given/new information. In P. Cole (Ed.), Radical prag-

matics (pp. 223-255). New York: Academic.Pulman, S. G. (1986). The recognition and interpretation of idioms. Unpublished manuscript,

University of Cambridge, Computer Laboratory, Cambridge, England.Quirk, R., Greenbaum, S., & Svartvik, J. (1985). A Comprehensive grammar of the English lan-

guage. London: Longman.Ruhl, C. E. (1989). On monosemy: A study in linguistic semantics. Albany: SUNY Press.Ruwet, N. (1983). Du bon usage des expressions idiomatiques [Concerning the correct use of

idioms]. Recherches Linguistiques, 11, 5-84.Swinney, D. A., & Cutler, A. (1979). The access and processing of idiomatic expressions. Journal

of Verbal Learning and Verbal Behavior, 18, 523-534.Tabossi, P., & Cacciari, C. (1988, July). Context effects in the comprehension of idioms. Paper

presented at the tenth annual conference of the Cognitive Science Society, Montreal.Wasow, T., Sag, I., & Nunberg, G. (1983). Idioms: An interim report. In S. Hattori & K. Inoue

(Eds.), Proceedings of the XIIlth International Congress of Linguistics (pp. 102-115). Tokyo:CIPL.

Wierzbicka, A. (1982). Why can you have a drink when you can't have an eat? Language, 58, 753-799.

Idiomaticity as a Reactionto L'Arbitraire du Signe inthe Universal Process ofSemeio-Genesis

Adam MakkaiUniversity of Illinois at Chicago, University of Hong Kong, andLACUS, Inc.

It generally is assumed by both linguists and lexicographers all over the worldthat DeSaussure's famous dictum identifying the relationship between le sign qie'(content) and le significant (expression) as basically "arbitrary" is correct.

The, principle of arbitrariness may be illustrated as follows: The concept ofdog and the word "dog" are in an arbitrary relationship to one another, becausedog, spelled backwards yields "god" and the concepts of dog and god are not thereverse of one another. Furthermore, the concept dog can be expressed as theFrench word chien, the German word Hund, the Russian word sobaka, andthe Hungarian word kutya, to mention just a few. Outsiders to any of theselanguages can easily be fooled by teaching them the wrong word. Thus, if you donot know Indonesian and are confronted with the choice between anjing andkuching (the one being cat and the other one being dog), you may have a hardtime choosing the right word. As it happens, anjing means dog and kuchingmeans cat in Indonesian, but in introductory linguistics courses students can beeasily fooled by asking them to choose between the word minum and susu for dogand cat, respectively. Some will vote for minum some for susu, although theformer means drink and the latter means milk. Kuching minum susu means thecat drinks milk.

The principle of arbitrariness was first formally worked out by DeSaussurewho, in his Cours de Linguistique Generale (1915), defined the linguistics signas consisting of a concept (the signified) and a string of sounds (the signifier)whose relationship was essentially arbitrary. Human societies get used to thesearbitrary designations of concepts by sound strings and hand this down to theirchildren. Insiders to human societies take these concept-sound relationships forgranted, whereas foreigners and children tend to question them.

297

298 MAKKAI

In this chapter I argue that the concept of 1'arbitraire du signe is a historicalproduct in the evolution of mankind's universal sign-producing activities and thatas such it had a beginning, a peak period, and a period of decadence. Theevidence for this thinking comes from publications such as Koestler's influentialbook The Ghost in the Machine (1967), in which brain evolution is traced, andM. A. K. Halliday's (1975a) classic on children's language acquisition, LearningHow to Mean, in which it is convincingly demonstrated that children, as theyacquire speech, seem to move through the main phases of human speech devel-opment in general. This, in turn, seems tro be a logical extension of Haeckel'stheory which, ever since its first formulation in the 19th century, maintained thatontogeny recapitulates philogeny.

But let us see in detail how such a concept relates to the problem of l'ar-bitraire du signe in general and idiomaticity in particular.

ON THE NATURE OF LINGUISTIC SIGNS

From "Martian" and "Venusian" to "Tellurian"

Primitive humanity (Neanderthals and their predecessors) are thought to havehad no more than 250 to 300 distinct noises to communicate with; these arebelieved to have corresponded to roughly 250 to 300 concepts in their dailyroutines such as signs of warning of danger, cries of pain and joy, indications ofthirst, hunger, and the mating urge, signs and sounds of territorial defense andaggression, and so forth. It is believed--and there may be room for doubt ordisagreement-that this state of affairs is roughly comparable, mutatis mutandis,to the holophrastic phase in the language development of contemporary children.Neanderthal speech is correlated by anthropologists and brain researchers astraced in Koestler (1967:129) with a stage of human brain development beforethe "mushrooming out of the human neo-cortex" (p. 129), which is much closerto us in time-35,000 to 40,000 years--and is correctable with the skull mea-surements of Cro-Magnon man. Depending on the anthropological and archae-ological sources one reads, Neanderthal man and Neanderthal speech can bethought of as I million years old. I am anxious to point out that I wish to take nosides in any debate related to "evolutionism" versus "creationism" as they areirrelevant to this discussion.

If the human voice, in terms of pitch, rhythm, intonation, and tempo, wascapable of being a manifestation of sensations such as hunger, pain, joy, danger,playfulness, grooming, and so on, and if the imaginable or reconstructable socialstructures associatable with Neanderthals were steady over long stretches oftime, this type of communication, although lacking "logic" in the modern math-ematical sense, must have been rather "logical"; after all there was not muchroom for misunderstanding. If a fairly limited set of concepts (250 to 300) is

13. IDIOMATICITY AND L"ARBITRAIRE DU SIGNE

C(oncept)1 c1

S Ivmbol / 00

C4 C5 C6It I&ound)l SS2 SS3 SS4 SS5 SS6

FIG. 13.1. Neanderthal communication, computer communication,Martian communication.

related reasonably steadfastly to 250 to 300 distinct signs manifesting in simplenoises, the communication system so reconstructed is not really basically differ-ent from modem computer languages, as shown in Fig. 13.1.

In modem parlance one might say that Fig. 13.1 shows a system characterizedby direct realization. There is no synonymy and no homonymy in this system asthere is also no composite realization, diversification, or neutralization. Forthose phenomena to arise we must look at another group of humans, much closerto us in time. Before we do that in order to characterize the languages of modemhumanity, let us ask the question: What would be the logical opposite of this"Neanderthal-Martian-Computer Language" system, in which one concept al-ways leads to one and the very same sound (or symbol)? It would have to be asystem in which any concept can be related to any sound symbol available to thespeakers. It would look as in Fig. 13.2.

*As can be seen by comparing Figs. 13.1 and 13.2, Fig. 13.2 is the logicalopposite of Fig. 13.1. In Fig. 13.1 each concept is expressed exclusively by itscorresponding sound-symbol; in Fig. 13.2 all of the concepts can lead to all ofthe sound symbols. How realistic is this in modem terms?, we might ask. Theanswer is that Fig. 13.1 is commonplace: The payroll of a large firm done on acomputer uses language very much like that depicted in Fig. 13.1. Computers areinanimate machines that lack human consciousness; they take nothing forgranted. This kind of language is used in science classes, during the instructionof mathematics, in usage instruction manuals, and so forth. Merchandise inmodem supermarkets is bar coded for the convenience of electronic pricing and

FIG. 13.2. Telepathic communication, poetic communication, Venu-sian communication.

299

300 MAKKAI

checking out; the bar codes always and unambiguously differentiate a can of beansoup from a can of mushroom soup and indicate the correct price.

But just as man does not live by bread alone, neither do we manage to use the"language of logic" all the time. Ever since the first lover called his lady a"flower," Venusian communication has been with us and it just will not go away.We, as a species, no sooner invented the convenience of direct realization inlogical communication as we found it constitutionally necessary to rebel againstit and to turn to the device of phoricity. Derived from the known term metaphor,phoricity refers to the feature of human languages whereby signs point not onlyat their references but at other signs as well and through these at a series ofnetworks of interrelated references. Modern human languages are so engineeredthat they always bring something "down," "over," "across," "forward," or"backward"; hence we have the phenomena of the metaphor, anaphora, andcataphora-to mention for now only the commonly recognized types.

Now modern Earth languages, or "Tellurian," to borrow John Algeo's term(1970), exhibit a combination of the communication systems diagramed in Figs.13.t and 13.2, shown in Fig. 13.3.

In the type of modern communication system outlined in Fig. 13.3, we have afair amount of direct realization as in the words "the square root of 16 is 4" and"may I have a pound of granulated sugar, please," but we also have a great dealof neutralization, as in the case of the English syllable well, which is a noun inthe sentence "the farmer dug a well," a verb in "tears welled up in my eyes," anadjective in "yesterday I was sick but I am well today," an adverb in "Johnspeaks Mandarin well," and an interjection in "well, I think so."

When we notice that a concept such as c[male parent] can be expressed inEnglish as father, which is neutral, as Dad, which is affectionate, as Daddywhich is affectionate and diminutive, as old man which is slang, we have en-countered the phenomenon of diversification.

Our human languages today, then, are characterized by the fact that besides alarge amount of direct realization they show neutralization and diversification tobe universally widespread features.

One of the most important realizaticonal relationships for our purposes iscomposite realization. The notion is familiar enough: Major sentence types an-swering the S (sentence) > NP (noun phrase) + VP (verb phrase) formulamanifest in the composite realization mode; sentences such as "dogs bark,"

C1 C2

SI 2

4 Cs .C6

I- I

I SS , A I;4 S5 S6

FIG. 13.3. Modern human communication: Earth language or Tel-lurian.

13. IDIOMATICITY AND L'ARBITRAIRE DU SIGNE 301

"cats meow," and "children play loudly" are all "compositely realized" insurface syntax due to the linear display of subject, verb, adverb, and so forth. Interms of signs composite realization works as follows: The sign elephant is asimple sign in direct realization to the concept c[large African or Indianpachyderm mammal with a trunk], and so is the sign white to the concept C[lightfrequency reflecting all of the sun's rays']. If one encounters an elephant that isan albino, one therefore may call it a white elephant. The manifestation mode ofthe expression will be composite in surface syntax, but the expression will not bean instance of composite realization but of direct realization, because the creaturethat is an elephant is of the white color; hence the words white and elephant dorealize an "elephant" and the "white color," respectively. However, as time goeson, societies become more complex and certain historic events take place thatleave an imprint of internal borrowing on natural languages. (cf. Halliday1975b, Lamb, 1966).

The King of Siam is said to have disliked a courtier once, and so, in order topunish him, gave him a white elephant as a gift. One must realize, of course, thatin ancient Thailand the white elephant was considered a holy animal somewhatas cows still are considered in India today. To kill or to neglect a white elephantwas considered a capital crime. Yet to care for a white elephant properly meantfinancial ruin. Because the white elephant was a gift from the King, it could notbe refused.

So what does white elephant mean today in spoken North American English? Itstands for "unwanted property, such as a large house, difficult to manage or tosell." It helps to know the story of the King of Siam and the courtier who waspunished with the expensive gift, but it is strictly speaking not necessary; there aremany fluent speakers of American English who use the expression white elephantin the aforementioned sense as used by real estate people who never heard of thestory from ancient Thailand. One must, as semanticist-lexicographer, treat whiteelephant as a separate entry in the lexicon of English from both white and elephantwith a definition all its own, that of "unmanageable property." This, to be sure, isa complex bundle of ideas-indeed entire networks are needed to express what isinvolved in the notions of "manage," "un," "-able," and "property"; yet thewhole designation of "unmanageable property" can be seen as a unitary con-glomerate, or Gestalt of meaning, hence a macrosememe or a definable unit ofmeaning, rendered familiar by social use and acceptance.

An Introductory Example

What kind of a sign, then, is white elephant in terms of our sign types depicted inFigs. 13.1, 13.2, and 13.3? It can be diagramed in simplified form as in Fig.13.4.

As can be seen from Fig. 13.4, the complex new expression white elephantmeaning "property difficult to manage or sell" has not lost its connections to the

302 MAKKAI

To z-alIav

To Morphologyand phonology To Morphology \

and phonology M

LEGEND:C ' 'concept'E 'expression'TRIANGLE: Composite unit

'a' followed by 'b'COATHANGER: Movable unit; can be examined 'alone' OR as part of a larger unit

FIG. 13.4. White elephant as an idiom and its place in a line-up ofconcepts and expressions.

concepts "white" and "elephant." That this is so is easily proven: Children andforeign learners of English, who are not familiar with a given idiom, try to workout its meaning by the traditional route of Fig. 13. 1; they tacitly assume that eachconcept has only one way of being expressed, and conversely, that each ex-pression leads to one and only one concept that is always the same. And this ishow children and foreigners are misled.

Permit me to illustrate from children's usage on some examples collected inthe United States:

When a child is naughty and the father wants to administer a spanking, hemay say, "I will fix your little red wagon," where the little red wagon refers tothe child's bottom and fix refers to the spanking. I heard children naively andinnocently reply to this, "but I don't have a little red wagon, Daddy!" -meaningquite literally that they had no toy wagon that was little and red. The fathersinvolved laughed and gave up the idea of the spanking; one picked the child up,spanked him while laughing, and said in an amused tone, "this is your 'little redwagon' [the bottom] and this is how I fix it." I imagine that this child will neverforget the meaning of the phrase "fix your little red wagon" (cf. Makkai, 1976).

The expression a monkey on my back in the United States or the correspondingBritish expression an albatross around my neck (with the latter being borrowed

13. IDIOMATICITY AND L'ARBITRAIRE DU SIGNED 303

from Coleridge's "Rime of the Ancient Mariner" and being therefore literary)both mean "an unpleasant, nagging problem, such as a complex of guilt, or debt,that will not go away." A mother once said in front of her child, "those Wednes-day afternoons at the hospital sure are a monkey on my back." What she meant tosay was that her volunteer work at the hospital is an unwelcome obligation shecannot get rid of. Her 4-year-old daughter replied, "But Mommy, there is noth-ing on your back, not a monkey, not a birdie-nothing!" The mother, who wasgreatly relieved by spontaneous laughter, explained to her child what a monkeyon one's back meant, and then went on to her hospital visit in a much improvedmood.

I was a recently arrived refugee immigrant in the United States when, shortlybefore the end of my 6th month in Boston, someone said, "You can't be serious,you're pulling my leg." I was startled and in broken phrases apologized, "pardonme, Sir, I did not mean to touch your foot." Amidst general hilarity I thus got myfirst practical lesson in English idiomaticity; I learned that to pull one's leg meansto "tease someone." Only years later did I find possible historical explanationsas to why and how the phrase probably arose-as it happens through the mercykilling of hanged prisoners who were suffocating: Pulling their legs broke theirspinal cords and ended their misery faster.

Composite Realization

White elephant, then, is an instance of composite realization, where the indi-vidual constituents of the composite form have some secondary connections tothe meanings of the constituents now suppressed in the composite form, butfreely in the foreground when the lexemes white and elephant are used outsidethe given construction referring to "unwanted or unmanageable property."

If one bears in mind the story from ancient Thailand, the designating of"unwanted or unmanageable property" as white elephant begins to seem lessarbitrary. To illustrate what arbitrary might be like in this connection, imaginethat "unwanted or unmanageable property" is called woopy-woop or toopy-toop.These are pronounced easily in English and presently unused and carry nomeaning whatsoever.

The language has opted for two familiar words in conjunction in a newcontext; one might argue, not because of the story from ancient Thailand, butbecause elephants are big as are houses, and a white elephant is rare; hence theimage of a white elephant is not a bad tag for a large house that is odd or unusual.This sounds realistic enough, yet fails to be convincing, because "unwanted orunmanageable" property also could have been compared to any number of largeobjects of an unusual nature such as a possible golden whale, but no such formactually is on record whereas white elephant is.

THE CONCEPT OF IDIOMATICITY-A MOREFORMAL APPROACH

It is now time to discuss the concept of idiomaticity more formally. The curvyarrows leading out of the concepts of "white" and "pachyderm" in Fig. 13.4lead us to physics and zoology, respectively; this is in the realm of generalsemantics or pragmatics, as some might prefer to call it. The dictionary meanings"white," "property difficult to manage or sell," and "elephant" are sememes orunits of meaning, which are, of course, like all units of meaning in every humanlanguage, semantically complex. The reason why they are units, nevertheless, isbecause in human memory they correlate with a unitary dictionary entry on theone hand and, on the other, contrast with their known meanings in isolation.White, on the lower level qua dictionary entry, is an adjective that can modify anoun; its meaning is its sememe, which can be expressed in a variety of stylescolloquially or scientifically or may be illustrated by a dash of white paint on adark surface. The entry elephant is a noun and as such can partake of thegrammatical behavior of English nouns referring to animate objects; its meaningis the SEMEME "African or Indian pachyderm with a trunk," which, again, canbe rendered colloquially, scientifically, on a drawing, or on a photograph. Whatinterests us particularly here is the complex expression white elephant. It must beseen as a single unit. It is a complex lexeme whose constituents white andelephant have preserved their historic connections to the sememes "white" and"elephant." Yet white elephant, which is a unitary noun, now has its ownprimary connecting line to the sememe "unwanted or unmanageable property,"and for modem speakers this is the really important line. The fact that whiteelephant (n) = "unwanted or unmanageable property" obviously must be spelledas two familiar words alphabetical under w and e in the dictionary, can be a sourceof confusion, psychologically speaking. Two words simply seem to be two wordsand nevertheless to have to think of them (in layman's terms) as one word presentsan uncomfortable contradiction, as if we were told that I = 2, the prime exampleof a falsehood we are taught to reject from early childhood onward. One certainlydoes not equal two arithmetically and so there can be no denying that theexpression white elephant (whatever it means) somehow contains two words.What we must come to tolerate here is that we are dealing not with prima faciephysical sight and identification but with a much subtler state of affairs, in order todo justice to this built-in mendacity in modem human languages that forces us tosee two as one and in order to introduce the main theme of this chapter, that of thegradual evolution of linguistic sign creation or semeio-genesis.

Pound's Poetic Typology Describes Word Formations

We need some terms in order to designate word creation in its various phases.The temptation arises to borrow three terms from poetic aesthetics, namely from

304


Ezra Pound (Eliot, 1963) who was, as is commonly known in literary circles,heavily influenced by the orientalist Ernest Fenollosa (Chisolm, 1963; Fenollosa,1920; Yip, 1969). 1 quote from Pound verbatim as given in Pound's essays editedby T. S. Eliot (1963):

That is to say, there are three "kinds of poetry":Melopoeia, wherein the words are charged, over and above their plain meaning,with some musical property, which directs the bearing or trend of that meaning.Phanopoeia, which is a casting of images upon the visual imagination.Logopoeia, "the dance of the intellect among words," that is to say, it employswords not only for their direct meaning, but it takes count in a special way of habitsof usage, of the context we expect to find with the word, its usual concomitants, ofits known acceptances, and of ironical play. It holds the aesthetic content which ispeculiarly the domain of verbal manifestation, and cannot possibly be contained inplastic or in music. It is the latest come, and perhaps the most tricky and undepend-able mode.

The melopoeia can be appreciated by a foreigner with a sensitive ear, eventhough he be ignorant of the language in which the poem is written. It is practicallyimpossible to transfer or translate it from one language to another, save perhaps bydivine accident, and for half a line at a time.

Phanopoeia can, on the other hand, be translated almost, or wholly intact. Whenit is good enough, it is practically impossible for the translator to destroy it save byvery crass bungling, and the neglect of perfectly well-known and formulative rules.

Logopoeia does not translate; though the attitude of mind it expresses may passthrough a paraphrase (pp. 19-20).

In order to properly distinguish my present use of the terms from Pound's andFenollosa's aesthetics, I use the term logopoeia in its etymological sense as"word creation" from the Greek logos ("word") and poein ("to make"). Insteadof the Poundian melopoeia for "sound poetry," I use the term onomatopoeia,more familiar in linguistics for "sound imitating words," and for the idiomatichabit of putting new signs together via an image, I use the hybrid Latino-Greekterm imagopoeia. There is one additional term that we need, referring to "lan-guage via gesturing," and this I call kinetopoeia. My indebtedness to Pound andFenallosa is very great indeed and my tampering with their designations for typesof poetry is in no way contradicted or detracted from by my borrowing from theirword-coining habits in order to introduce my concept of idiomaticity via image-based word formation; that is, imagopoetic logopoeia, for that is what idi-omaticity is really all about.

Generally humanity, in creating signs, is always engaged in some form orother of semeio-genesis via logopoeia, and the kind of logopoeia we discover atvarious levels is characteristic of the stage of development of the language groupin question. But how does this all take place?

I suggest in this article that roughly at a time when the human neo-cortexbegins to "balloon out" as suggested in Koestler's brilliant book on brain devel-

306 MAKKAI

opment (1967) and Neanderthal language yields its place to the much richer andmore complex languages of Cro-Magnon man, we abandon the elementaryphonetopoetic-onomatopoetic way to verbal creation and embark on the moremodem course of imagopoetic internal borrowing.

The question arises: What are the probable reasons for abandoning meaningdesignation by the allocation of sounds for the device of borrowing an image?The answer probably lies in the discrepancy between the available and distin-guishable sounds and the number of new concepts humans need for survival. TheOxford English Dictionary (OED) has almost 500,000 entries in it (lexemes);these in turn, require minimally 10 to 20 words (lexemes) to define these entries.The OED, therefore, contains minimally 5,000,000 but more likely 10,000,000individual sememes, certain well-chosen configurations of which constitute defi-nitional macrosememes. To illustrate: If the entry dog (n) is defined as "domesti-cated canine mammal of the canidae species," the words between the quotationmarks are the definitional macrosememe and the constituents therein are se-memes inside of it such that each one correlates to a lexeme and, therefore, canbe looked up in turn. (It happens not infrequently that one has to look up themeanings of words one finds in a definition before one can make sense of thedefinition in itself.) All definitions are intrinsically highly complex and lead toencyclopaedic knowledge of various disciplines. The notions of "domesticated,""canine," and "mammal" all can be studied in extenso so that one has to have anentire library at one's disposal. It is therefore very difficult to estimate thenumber of pragmo-semantic linkages that one can reach from 5 million to 10million sememes, but a conservative multiplication by a factor of 10 seems notunreasonable. We can imagine informally, therefore, that the OED shows anupper line-up as follows:

1. Number of entries: half a million.2. Number of definitions (macrosememes): half a million.

3. Number of sememes: 5 million to 10 million.4. Number of pragmo-semantic linkages: 50 million to 100 million

Yet all of this information is encapsulated in far fewer morphemes and only afew dozen phonemes. This would be the OED's lower line-up:

1. Number of morphemes: A few thousand.2. Number of phonemes: A few dozen.

Let us think of a class of 30 students. Their individual biographies, and let usimagine that they have to share one single typewriter to write down their con-densed biographies on one page each. The phrase "I was born in X" probablywould appear in all 32 biographies; some students will be born in the same city

13. IDIOMATICITY AND L'ARBITRAIRE DU SIGNE

(X), others elsewhere; the year may be the same or not; the months and days willdiffer, but there might be an occasional pair of "birthday twins." Each biographywill constitute an individual text; after all, these are separate individuals. Yeteach text will have a great recurrence of similar or identical phrases, the wordswill be even closer and, given the fact that they had to use the same typewriter,they will have used the same limited set of letters to express their individual lifehistories. If one of the students wants to be "original" and wants to say how hardhe intends to work after graduation, he will be unlikely to invent a new word andwrite "I shall be working *brummmmmmmmmmmmmmmmmmm"; on theother hand, he might try to say something like "I shall be working more dili-gently than seven Vietnamese shopkeepers, seven Polish farmers, and sevenChinese restaurant owners put together."

The student who wrote *brummmmmmmmmmmm, to express the notion of"very hard work" engaged in phonopoetic logopoeia. The student, however,who promised to work harder than seven Vietnamese, seven Poles, and sevenChinamen put together is availing himself of the possibility of imagopoeticlogopoeia. Everybody knows that the Vietnamese, the Poles, and the Chinese aresome of the most diligent immigrants in the United States and one easily canvisualize the restaurateur, the farmer, and the shopkeeper at work. Multiplyingtheir effects by seven is, of course, a difficult calculation to perform exactly, butresorting to such numerical hyperbole is, nevertheless, a possible way to expressthe notion of "to a high degree."

The instance just cited is imaginary, and even if it did happen would notamount to anything we could lexicalize in a dictionary; it would amount to whatHockett (1958) called a "nonce form" without social significance.

The Birth of a Famous American Idiom

Next, I would like to illustrate the imagopoetic creation of the idiomatic lexemered herring.

In the 19th century, fugitives from justice on the run, when pursued by blood-hounds with a sharp sense of smell, would try to escape by rubbing pieces ofherring on low-lying branches of the trail in order to confuse the sense of smell ofthe hounds. Herring tends to be reddish in color and, when old, has a foul smell.As American civilization developed during the course of the 19th century andlegal issues arose, people found it necessary to make quick and handy referenceto false or phony issues. They just could have said, of course, "this is afalse/phony issue" and no doubt many did. But people, whether in law enforce-ment or private life, who were familiar with the habit of escapees who had goneto the West Coast (particularly Oregon where this phrase is said to have origi-nated) knew about the habit of rubbing "red herring" on branches along escapetrails and began to use the phrase red herring in the sense of "confus-ing/phony/false issue."

307

308 MAKKAI

One who does not know the meaning of the idiom red herring is forced by theexpression to think of the color "red" and of the fish called "herring"; thereason, of course, is that the hearer already knows those terms but cannot makesense of them in combination. As time went on, the phrase became so popularthat nowadays all dictionaries carry it. As shown in Fig. 13.5, its structure is thesame as that of white elephant, adjective + noun, with the resultant combinationfunctioning as a new noun with a nondeduceable, hence unitary meaning (amacrosememe), which, of course, is polysemantic in cognition and the pragmat-ics of the "real world." There is a subtle but real difference in the degree oftransparency of the two expressions: White elephant, according to many speak-ers, is easier to identify with an unsellable building than red herring with a phonyissue. Characteristically, white elephant refers to an object-hence the relativesimplicity in visualization-whereas red herring refers to people's verbal behav-

/to physics (To logic, philo (To ichthyology)

GENERAL SEMANTICS law, the structure of GENERAL SEMANTICSargumentation, valid

I evidence, lies.)C. JV C/-_ /<CMM/.monev issue/ (SEMEME) /herring/ (SEMEME)

(LEXEME)Noun

T To morphologyTo morphology and and phonologyphonology >

LEGEND The shaded triangle in the middle is the contemporary nominal lexeme red herringwith the contemporary meaning 'phoney issue'. Its two constituents red and herring, res-pectively, can be looked up in a dictionary separately and arrive at physics on account ofthe color 'red' and ichthyology on account of the fish 'herring'. The idiomatic weaning'phoney issue' can lead the investigator to logic, philosophy, law. the syllogism. the struct-ure of argumentation, lies and the subterfuge. All of this is SINULTANEOUSLY AVAILABLE orIGNORABLE with concentration on what is necessary at the moment. If red herring were aChinese idioms written with the characters for 'red' and 'herring'. the etymology of thisexpression would be outwardly visible as it is in the case of 'quickly-fast-imediately'which is written by the characters for 'horse' and 'back'.

FIG. 13.5. Red herring as bimorphemic lexeme.

iMUUI..T

LE XOLOY

13. IDIOMATICITYANDL'ARBITRAIREDUSIGNE 309

ior, their habit of not talking honestly or straightforwardly about an issue. If onewanted to do better justice to the subject, one would have to classify whiteelephant as an instance of concretelobjective imagopoeia and red herring asabstractlremoved imagopoeia. This would further depend on the obviousness orobscurity of the diachronic information available on any given form. What allsuch idiom formation shares in common is this: instead of creating a new wordvia sound to designate a new concept, the speech community involved borrowstwo already existing terms and recombines them in a new sense in such a waythat the old senses of the constituent parts remain perceivably present; it is theoriginal "literal sense" of the combination whose historic appropriateness occa-sions the choice in the first place. The result is an extremely complex gradient ofimages ranging from the completely transparent via the semitranslucent to thecompletely obscure.

In what follows I try to illustrate the concept of idiom formation on some lessknown Chinese examples and explore how these compare with some An-glo/American idioms. After this I examine the validity of the claims advanced byCharles E. Ruhl who asserted that there are practically no idioms at all-whatpeople consider to be idioms are really only cases of "pragmatic metonymy." Iconclude in agreement with Lamb who considered Ruhl's position to be a case of"either-or thinking" when, in fact, in order to come to a proper understanding ofidiomaticity, one must tolerate the discomfort of double-vision, or practice"both-and thinking."

IDIOM FORMATION

The Chinese language offers some of the best examples of idiom formation froman unrelated non-Indo-European language for the following reasons: It has aunique writing system that is both phonopoetic and imagopoetic in its own right,and it is a language with relatively few segmental syllables for which Chinesecompensates with a rich tone system (Mandarin has four tones and Cantonesenine). In my characterization of Chinese idioms I rely heavily on Rohsenow(1984).

A Brief Look at Chinese Character Writing

The Chinese language, like all other languages is primarily spoken; in thePeople's Republic of China today at least half of the population of 1.1 billionpeople is still illiterate. The writing system of Chinese is unique among theworld's languages. Rohsenow (1984) wrote:

To summarize the history of the development of Chinese characters briefly, theearliest examples of Chinese "writing" from approximately 2000 B.c. are still

310 MAKKAI

somewhat obscure, but by the period of the "Oracle Bone" inscriptions (1400-1200 B.C.) the writing system had already basically reached the form in which weknow it today. Chinese characters are traditionally divided into six basic classes("lift shui") on the basis of the manner in which they are formed. This division wasfirst made in the classic Zhou Li ("The Riles of the State of Zhou") about 300 B.C.

These six classes are:

1. Simple Pictographs:A< (a picture of a) person, originally /ren/ )El (a picture of) the sun, originally I!ri/f (a picture of) the moon, originally /yue/ )* (a picture of a) tree, originally /m/ Xh (a picture of) a horse, originally /ma/ >

2. Ideograms:* (person with arms extended) = 'big/great/large' MdA/

-(one line) = the number one Iyx(two lines) = the number two pert(three lines) = the number three /san/

J (vertical above line) = 'up/on' /shAng/ 'above'TK (vertical below line) = 'down/below' /xia/ 'below'

3. 'Met Meanings':E 'sun' + q 'moon' = B 'bright' lming/

4. Repeated Characters:* 'tree' (doubled) = ; 'woods' /lin/*'tree' (tripled) = 'forest' /sEn/

5. (Internal) Phonological Borrowings'wheat', phonologically reconstructed as */aijk/ comes to mean 'come',

now written M

6. 'Picto-Phonetic Characters':N /ma/from * 'female' + A /md/'horse' = (mother), /ma/from rl ri 2 mouths' + Eg /ma/'horse' = 'to scold'

(pp. 335-336)

As we can see from this brief sketch, C'hinese writing at the beginning was anattempt at iconic representation of observed reality. This gave way to the ide-ograms, which are posticonic and presymnbolic, as it were, perhaps best charac-terizable as indexical in nature. The categories of met meanings and repeatedcharacters are a further extension of the indexical phase, but with internalphonological borrowing we enter the symbolic phrase, which comes to fullfruition in the picto-phonetic characters, which, therefore, are also the mostidiomatic ones of all Chinese characters.

The picto-phonetic character says to the learner something like this: Think ofthe segmental sounds Im/ and /a/ as in "horse" (pronounced lmrAl on a risingtone) but pronounce it on a contrastive high tone as ImRI, and to remind you ofwhat it means, write "woman" before it and let this now stand for "mother."

The second picto-phonetic character means "scold"; it says to the learnersomething such as: Think of the segmental sounds Im! and /a/ as in ma


("horse"), but pronounce them on a contrasting tone so as to differ from both"horse" and "mother" and then to remind yourself of what you mean write "twomouths" on top of it and let the new composite character stand for "scold."

The large majority of Chinese characters are composed this way today and sothere is reason to characterize the Chinese writing system as the grapho-log-opoetic activity of combining phonopoetically evolved, preexisting lexemes with(a) pictographs, (b) ideograms, (c) their met and repeated extensions, (d) internalphonological borrowings, and finally (e) the mixing of a phonetically predomi-nant character with an imagopoetically predominant character.

Chinese writing, then, can be said to be multiply idiomatic and its line ofevolution can be characterized as being in accordance with the expected line ofdevelopment moving from the "concrete" toward the "abstract" (cf. Fabricius-Kovacs, 1961; Kronasser, 1952; Makkai, 1972).

Some Better Known Chinese Idioms

Let us take the two characters b md ("horse") and ± ("up/on") as two famil-iar characters we encountered earlier and ask ourselves as outsiders to the Chi-nese language what the two possibly could mean if put together in combination?First of all, we must ascertain that the combination md shting is possible inChinese, so we propose it to a reliable native speaker. The answer is affirmative;the combination does occur and is written by the two signs as expected %h±. Wenow can rethink the formations of these two characters in trying to guess theircombined meaning: The character for "horse" was iconic in origin; the changesin the development of the simplified picture of a horse to the classical character(still showing "four legs") to the simplified character (replacing the "four legs"with a simple line) iy can be seen as an imagopoetic process. The fact that theconcept "horse" already had been related to the syllable ma has to be seen asgiven; this relationship arose prior to writing and was the result of phonopoeticlogopoeia, that is, the creating of words via symbolization by means of sound.

The second character A stands for "up/on"; it can be characterized as ide-ogram; as such it is more indexical than iconic. Because the up and downrelationships do contrast in terms of their positions relative to the horizontalstroke, this idiogram is imagopoetically derived to a significant extent; the imageis indicative of a physical relationship because the characters do not mean "exal-ted" versus "humble" or "rich" versus "poor," which they also have beenstretched to mean, as in J« gbiM "people on the top socially."

We thus are dealing with two signs of diverse origin but concrete meaning inthe contemporary sense "horse" and "up/on/back" respectively and this, inturn, gives us every reason to expect that the meaning of the combination shouldmean something like "the back of a/the horse" or "horseback." When this issuggested to Chinese speakers, however, they show great hilarity and say "yes, itcould mean that, but it really means 'fast, rapidly, right away, immediately'."And so we encounter yet another phase far removed from the ancient way of

312 MAKKAI

phonopoetic logopoeia, when people could have made up any suggestive soundthat for them somehow symbolized the idea of speed. They could have said/rrr=rrrrrrr/, for instance. The midcentral approximative I/r, common in Man-darin and reminiscent of the American English pronunciation of err, could havebeen perceived by speakers of ancient Chinese as a symbol of "speed" or"haste"-the sound, especially if sustained, may have a ring or "urgency"about it-but this did not happen. Rather the Chinese language evolved in such away that it selected two already established and, of course, arbitrary signs, thatfor "horse" and that for "back." These., in turn, were put together to combine ina new way to become the carriers of the concept of "quickly" or "fast" (cf.Hockett, 1958).

What has happened here? What is there or was there about "horseback" thatcould justify this combination as the expression for "quickly/immediately"?

Chinese society even today is hea-vily rural and agricultural. Whereas the20th-century conveniences of the telephone and telegraph have reached Beijingand the big cities, the fastest means of bringing news, letters, and parcels in thenot-too-distant past was primarily by riders on the horseback. Thus when aChinese speaker says today "pass me the pen, quickly!" he is really saying "passme the pen 'horseback'!" This is first surprising to the foreign learner, but aftersome thought it starts to make sense. We do have a rather similar construction inEnglish. Consider "come on let's go, step on it!" The object one is told to stepon is the accelerator, the gas pedal of the ubiquitous modem mode of transporta-tion, the familiar automobile, and the phrase is used when there is no automobilein sight.

Chinese md shing and American English step on it have this in common: Bothrefer to speed and both use an image from a common mode of non-pedestriantransportation. In the older Chinese example this happens to be horseback, and inthe case of the newer, more technologically orientated American English, it is theaccelerator of an automobile. This can be represented in diagram format as inFig. 13.6.

But if horseback was indeed the fastest commonly available mode of transpor-tation in old China and if, indeed, the American car is the most common meansof speedy transportation, then the choosing of these compound signs is no longerquite as arbitrary as was the choosing of the old, primary signs, such as Englishtree for "arboreal plant" and Chinese ma for "horse." As we saw, no soundsequence such as /Irrrrrrr/ was chosen to express the idea of fast either in Englishor in Chinese. The phonopoetic mechanisms of word creation (logopoeia) dimin-ish as civilization advances, and so the speakers, acting semiconsciously, start toborrow already existing signs and to recombine those to express the meaning athand. The first time someone in ancient China said "Md shdng!" and the firsttime an American said "Step on it!" the listeners must have felt that these nonce-forms (cf. Hockett, 1958) were memorable and appropriate, and people started,equally semiconsciously, to imitate the initiators. Today md shdng in Chinese and


The Concep

f'speedy",

{traditional ChineseTransport taion

f'horse' ma gas pedal' = it

'back' = sh~ng -accelerate' = step on

FIG. 13.6. A Chinese and an American English idiom, both using im-ages from a common mode of nonpedestrian transportation.

step on it in American English are lexical facts the linguist-lexicographer mustaccount for.

The Chinese expression for "danger" is xido xin; it is written by the charac-ters iJX{.'; the literal meaning of compound is "small heart," an interestingexample for our consideration. The human heart is invisible to the eye, but fearand anxiety can cause a sensation of narrowing in the chest. In English onespeaks of a "sunken heart" when one is shocked or frightened.

In modem times, as Western technological inventions reached China, thechoice arose whether to create new native expressions for these commonly usedgoods or to name them in some other way. Consider the telephone. We in theWest take this as a single concept and seldom analyze it; one has to go toetymological studies and Greek to see telephone as a complex sign made up oftelos ("goal, distance") and phone ("sound"). If understood etymologically,telephone means "distance sounder," a fact well echoed in its German nameFernsprecher ("distance talker"). Yet when in the 1960s in Kuala Lumpur,Malaysia, the National Language Institute (Dewan Bahasa Dan Pustaka) weregiven the suggestion to call the telephone bunyi jauh (from Malay "distance"and "sound"), the suggestion was rejected and they decided to use talipuninstead with the customary substitution of /p/ for Ifl. Not so the Chinese. Theycall the telephone tidn hua, written'jr; the characters stand for "electric" and"talk/speed." The idea of sound is not expressed, but electricity, a primarynotion in Chinese, is favored in a number of similar expressions.

When computers were introduced in China, a similarly ingenious idiom arose:To "electric" as in tien hua ("telephone"), they now added the character for

313

1

314 MAKKAI

"brains": The result is tiean ndo, written J ("computer"), literally "electricbrains." The third member of this triad is tidn ytng, writtentr(("electric shad-ow/image"), which stands for "motion picture, movies."

ARE IDIOMS CASESOF PRAGMATIC METONYMY?

In Western linguistic speculation (Ruh], 1989) there has been a recent trendtrying to question the existence of idioms and explaining as many idioms aspossible as cases of pragmatic metonymy. Ruhl (1974, 1975, 1976, 1978, 1979,1980, 1982, 1989) reiterated on numerous occasions that idioms exist only in theheads of foreigners and that mature native speakers always know the literalmeaning that is the base of a given idiom. This notion, attractive to many, ischallenging to the semanticist-lexicographer working with idioms and so I feelobligated to devote attention to Ruhl's arguments against idiomaticity.

The well-known expression kick the bucket ("to die") is not an idiom inRuhl's scheme of affairs, because of the following considerations: Kick ("rapidmotion with the foot resulting in hard contact") is a sign of incipient life anddeath both, in general. Thus the first sign of an embryo's continued healthygrowth inside the womb is when it first starts to kick, usually in the 5th month ofpregnancy. When a cold engine in a car first refuses to start and requires repeatedeffort either by start-key and battery or by hand cranking and then finally arts; jtis said to kick in, to "start coming alive." The phrase to kick in is now soextended as to cover the meaning "take effect" in general; one hears statementssuch as "My salary kicks in after the 28th of each month." To signal the end oflife, kick over is used, as in "Mrs. O'Leary's old cow kicked over this morning."

Ruhl (1982) had no explanation for the bucket in kick the bucket and so headmitted that "there is something idiomatic" (p. 304) about the expression. Thetruth is that nobody really knows where bucket comes from in this phrase. Somepeople think that it is in memory of Western-style hangings in which the criminalis stood up on a bucket with the noose in his neck and the rope looped around thebranch of a tree; then the sheriff comes and kicks the bucket out from under thecriminal whereby he is hanged. According to Flavell (1972), bucket comes fromthe French bucket, which meant "beam." According to Flavell, the phrase origi-nated in 16th- and 17th-century English farmhouses where pigs were slaughteredwith their feet being tied to the beam, or "bucket." The pigs as their last sign ofreflex were "kicking the beams" to which their feet were tied. Whereas it makessense for such a widely spread idiom to have originated in England, the questionremains as to why and how French bucket, now dead in English, predominatesover the common word beam?

Ruhl's (1982) assertion that kick the bucket is not an idiom has the partial


merit of calling attention to kicking as a sign of last reflex and incipient life, butthe acceptance of these side associations heavily depends on diachronic informa-tion (as in the case of the story in England of old-fashioned pig slaughter) and onall sorts of other cultural information beyond the scope of work for the descrip-tive lexicographer. As Bolinger (1976) convincingly argued, idioms have outsiderelations, and the associative links that parts of idioms have with other lexemes inthe language rest on diachronic awareness.

Ruhl (1978, 1980) devoted a separate article to to each of the followingexpressions hit the sack ("go to sleep") and break the ice ("start talking"); hisintention was to explain that these are not idioms but metonymic extensions ofthe underlying common senses of hit, break, and ice.

The hit of hit the sack also occurs, Ruhl (1978) convincingly demonstrated, inhit the deck ("lie down suddenly") and call me the moment you hit town meaning"when you arrive in town." From these and other examples collected frommodem American fiction, Ruh] concluded that hit has the meaning "make sud-den contact with." Because sack is used in slang to mean "bed" as in "John is asack-rat" ("he sleeps a lot"), Ruhl suggested that hit the sack is, in fact, under-standable as the combination of "to make sudden contact with" and "informalplace to sleep."

Break the ice is similarly dismissed as an idiom. Ruhl (1980) demonstratedthat break means "suddenly to disjoin, to dissociate" and also "suddenly torelease" as in "Finally my fever broke" and "The news broke around 8o'clock." The word ice can be shown to enjoy widespread use referring toisolation, frozenness, sadness, and so forth, as one talks about the icy look onesomeone's face, and so on. To break the ice, then, Ruhl concluded, is not anidiom but a metonymic sense extension, which understandably says "make thefrozen silence/sadness go away."

If the concepts "go to bed quickly" and "start talking" were expressed as/hummmmmmmmm/ and /wheeeeeee/, respectively, the question whether theyare idioms or not never would arise; speakers would accept them as pho-nopoetically created, arbitrary signs of ancient origin. Because, however, in thelate 20th century we have these two compound expressions with a familiarsyntactic pattern (to verb [V] the noun [N]), speculation naturally is arousedregarding their origins and the question of just how arbitrary they are. Anydescriptive lexicographer, whether working on a bilingual dictionary of Englishor on an OED-style, monolingual defining English dictionary would be welladvised to include both hit the sack and break the ice under the lemmas for hitand break, if the task at hand is a description of contemporary English usage.Ruhl's point, however, that literal senses can be found in such expressions byeducated native speakers who also have a broad range of literary and culturalassociations is a valid one. We thus come to a crucial question in the analysis ofidioms: for whom is a given expression an idiom?

315

THE RELATIVITY OF IDIOM PERCEPTION

"For whom" is a dangerous and loaded question because it arouses the specter ofrelativism in linguistics, a discipline, which is experiencing a hard time acting asan objective science (Yngve, 1974, 1986). The question "for whom?" wouldhave been difficult to raise during the 1950s when the neo-Bloomfieldian struc-turalists were generally somewhat discouraging of meaning, and in transforma-tional generative grammar (based on C'homsky, 1965) "for whom?" was, onceagain, a nonquestion as the grammar worth describing was the "competence ofthe ideal hearer-speaker in a homogenous society," a situation that does not existin reality and that had to be artificially created by Chomsky in order to to makelanguage fit the Procrustean bed of symbolic logic. Yet with the 1960s therecame urban sociolinguistics (Labov, 1964, 1976) with a resultant more or lessstandard acceptance of variationism. In grammar (mainly syntax), we are nolonger shocked when we hear "this is 'grammatical' in my speech, but it may be'ungrammatical' in yours." An utterance such as "I ain't got nuttin' to do withyou no more," once held up as the prime example of "illogical speech" wherenegations of negatives allegedly created unwanted assertions, is viewed calmlynowadays as an instance of "emphatic denial" in Appalachian speech and inBlack English. The question of "for whom?" should be, I think, no longer tabooin modem linguistics. Yet in lexical semantics, an important and sensitive areawhere tolerance must be practiced. the issue of "relative understanding" neverwas raised, with two notable exceptions, Welch (1974) and Lamb (1990). Theformer deals with the layman's and the Latin scholar's awareness of Englishpolysyllables of Latin origin; the latter warns against the hazards of either-orthinking.

The thrust of Lamb's (1990) argument is that whether hit the sack and breakthe ice are perceived as idioms or as metonymic extensions of literal phrasesdepends on the cognitive capacity of the user/analyst and the intended target-audience. The situation is paradoxical in the true Greek sense of paradox ("anapparent contradiction that displays a hidden truth"): The only realistic way toaccount for these phrases is to state that they are both literal and idiomatic. Thischapter then, is offered as an argument for the inclusiveness of both-and thinkingin contrast to the exclusiveness of either-or thinking.

"Regularity" can be found in break the ice and hit the sack where the gram-matical behavior of these forms under inflection is concerned: The past tense ishe broke the ice and never *he break-the-ice-ed; the past tense of hit the sack ishe hit the sack and never *he hit-the-sacked. The same is true about kick thebucket; the past form is he kicked the bucket and never *he kick-the-bucketed,which shows that the morpohemes break, hit, and kick remain in these ex-pressions with their customary morphological behavior whether or not they alsorefer to the acts of "breaking," "hitting," or "kicking." Consider as furtherexamples of the same principle English understand-understood and withstand-

316


withstood, which are never inflected as weak-verb forms *understanded and*withstanded, although physical "standing" is not necessarily involved in themental act of "comprehension" and the social act of "resistance." One makesthe best descriptive sense, therefore, by suggesting that the morpheme stand ispresent in these expressions but the lexeme stand ("be positioned erect on one'sfeet") is not.

What is required is that modem linguists give up, at long last, their uncon-scious adherance to the Bloomfieldian notion of the morpheme (Bloomfield,1933) as a "meaningful unit" (p. 160) where Bloomfield used the term sememefor the "meaning of a morpheme" (p. 160), and come to accept the existence ofthe intervening lexical stratum.

The recognition of lexemes is of potentially great benefit for linguistics be-cause the lexemic stratum can handle syntax in its tactical dimensions. Meaningand "deep structure" can be handled on the next stratum higher up, in semology,the place where the definitions of the lexemes, the actual entries in the dictionary,must be handled in any event.

But let us return to Ruhl's examples kick the bucket and hit the sack in order tosee how they could be used literally. Imagine a scene of target shooting: Amongthe targets to be hit there are some beer cans, some wooden dolls, and behindthem, covering the wall, an old sack. Max, a novice, keeps missing both thewooden dolls and the beer cans. The instructor says: "Max, stop hitting the sack!Aim at the bottles!"

On a narrow corridor in a boarding house, where the running water is shut offfor repairs, there are some buckets with water in them for people to wash in themorning. Max, who is unaware of the water shutdown and the water storagebuckets, moves around in the dark. Accidentally his foot comes into collisionwith a pail that makes a loud rattling noise. One of the boarders gets up to seewhat caused the noise and tells his roommate: "Max, the newcomer, kicked thebucket." The point is that no native speaker can say this without knowing that heis punning. In Stanley Kramer's motion picture, It's a Mad, Mad, Mad, MadWorld, actor Jimmy Durante, who fell out of a car in a hairpin curve, tells theonlookers where some hidden money can be found, and then his foot comes intoviolent collision with a pail that rolls down the hill before he dies. The audiencerewards this visual literalization of the idiom with uproarious laughter. No oneever denied, accordingly, that idioms do behave partially as if they were literalconstructions, but we must observe the native population in their daily interac-tions of how they actually use idioms.

Lamb (1990) has convincingly shown in terms of the formalism of relationalnetwork diagramming what I am saying here in plain prose: The questionwhether or not a form is an idiom or not is relative to the consciousness of theuser. Because some users are both native speakers and linguists-or even literaryscholars-the range of possible associations is rather wide.

Speakers of dialects other than American English when coming to the United

318 MAKKAI

States tend to acquire idioms such as 'He has two strikes against him" and"He'll never get to first base" entirely outside of the baseball-based sport contextwith which they are not familiar. The first one means "to have wasted twochances out of three after which one must stop trying" and the second means "toestablish oneself in a position of strategic importance from where further movesbecome possible." Only after prolonged stay in the United States do such naiveusers of baseball idioms discover where the phrases come from.

A SECOND LOOK AT CHINESE IDIOMS

But let us return to some of our Chinese examples used in this chapter earlier.Chances are that native speakers of Mandarin hardly ever speculate on what agiven character originally stood for; for Chinese people the characters simplyindicates the segmental phonemes /in/ and /a/, which must be pronounced withdifferent contrasting tones, depending on the meaning. With the rising tone mdmeans "horse," and in the written form no additional character is required toexpress the idea of horse. If the same segmentals are pronounced on a high-leveltone as md, the word means "mother," and the character for "woman" must beadded before the character for "horse." The question arises: Is the meaning"horse" at all present in Mandarin "mother" then? Of course not. Yet thecharacter for "horse" is there! What does that make the combined char-acter A ("mother") in Chinese? My suggestion is that it is an independentlexeme on the spoken level; after all it has its own tone, which is in contrast withthat of both "horse" and "scold." Graphically, however, it is an idiom that in the"horse" character calls attention to par: of the pronunciation (the segmentals)and in the "woman" character reminds users of a "human female." The fact thatthe combined character does not mean 'mare" (female horse) or "woman rider"or a "woman as big or strong as a horse," or any such possible combination,shows that the combined character is a grapho-lexemic idiom. The Chineselanguage is probably unique in having developed this kind of idiomaticity. Anative speaker of Chinese has no need to rationalize to himself or herself thecomposition of the character for motherei'" in this way, but it is said here in HongKong that it is a powerful mnemonic device for foreigners and for children. Wethus can understand why the question 'for whom?" is far from being irrelevant.

CONCLUSIONS

What theoretical conclusions can we possibly draw from all of this? Thereappears to be a universal constraint on all idiom formation and it is this: Whereasnonidioms probably arose through phonopoetic arbitrariness, idioms tend to arisevia imagopoetic arbitrariness, thereby amounting to a new phase in the evolution


of human logopoetic, that is word-building, activity. Imagopoetic arbitrarinessseems to be closer to the real world from the point of view of our contemporaryinclinations toward logic. Societies that already have developed a fair number ofsigns via the usual l'arbitraire du signe, where arbitrariness means the linkage ofa sound sequence to a macrosememe, when forming newer sets of concepts donot resort to the ancient phonopoetic arbitrariness, but start to engage in anactivity of a nonacoustic but visual order: They now select already existing signsand start recombining them in terms of a suggestive image. Hence the name ofthis kind of idiomatic logopoeia as imagopoeia, that via image-creating wordformation.

The Chinese idioms md shdng ("quickly"), tidn hud ("telephone"), and tidnndo ("computer") are all such imagopoetically created idioms whose structurewould indicate that they are younger formations in the history of the languagethan the primary signs that were created phonopoetically.

At the beginning of their lingual evolution ancient humans were, it wouldseem, more orientated toward sound than toward visual images. The fact thatwriting is so much younger than speech would seem to support such a line ofspeculation. Phonopoetic activity was, in all likelihood, onomatopoetic, thenotion of onomatopoeia, sound-imitating designations, is, in fact, the motivationfor the coining of terms phonopoetic-phonopoeia (sound), logopoetic-logopoeia(words), and imagopoetic-imagopoeia (images) as used in this chapter.

Also consider the fact that in ancient Greece, Homeric epics were memorizedand performed, with performance sometimes lasting a day or two without anywritten records. The invention of the hexameter by the Greeks certainly musthave been a help in the memorization process as syllable length, stress, and theopenness or closedness of the syllables involved correlated with the lexemes inquestion whose linear syntax, in turn, "told the story" of the Iliad and theOdyssey. As children grow up, they, too, show tendencies of having their memo-ries first rooted in touch, second in sound, and only around the age of 6 yearswhen elementary school starts to bring drawing and literacy, in visual images.

It is not difficult to see how the sound-based systems had to start breakingdown as the brains of social humans expanded cortically and developed furtherand further needs to build connections between concepts and expressions. Theonce more or less natural connection between pleasure and danger as notions andthe cries that expressed them, although remaining in the archaic residue of theonomatopoetic word stock of modern languages, gradually gave way to sapientmodes of logopoeia whose chief characteristic, by necessity, became De Saus-sure's l'arbitraire du signe. But as more and more such arbitrary connectionswere built and allowed to proliferate, speech communities reached a certainsaturation point. The resultant homonymic overload started to threaten socialusefulness, and speech communities began semiconsciously to rebel against theexclusive and total arbitrariness of the sign system. When speech communitiesreached their respective stages of homonymic overload, they also reached the

320 MAKKAI

phase of phonopoetic saturation. What this means is easy to see and to validate:Hardly ever do we manage to create a neologism in one of today's languagesbuilt on the archaic residue of phonopoetically anchored nonproductive patterns.The number of strong verbs of the sing-sang-sung type are limited and fixed inEnglish; newly made-up verbs are all weak and have their past tenses in -ed.Onomatopoetic words such as hiss, thud, bang, and meow are much more reluc-tant to sprout new buds in the lexicon of English, whereas the sciences andtechnology have no difficulty coming up with abstract new terms by cleverlyremobilizing a few Greek and Latin sterns, prefixes, and suffixes.

With neocortical expansion comes the phase of sapient logopoeia which,historically well before the invention of writing, starts using visual imagery. TheHomeric "standing epithets" of ox-eyed Hera and the rose-fingered dawn, em-bedded in the rhythmic repetition of the Homeric hexameters, were, in fact,visually based images.

What are some of the favorite sources of imagery in modern idiom formation?They seem to include parts of the body--nouns such as hands, arms, feet, legs,the head, the eyes, the ears, and so on, with "primary verbs," which after firstbeing used for some bodily function such as break, kick, and take, may becomethe subjects of Ruhl's metonymic extension principle.

Another heavily favored area is that of animals and familiar elements ofnature. The notion of slipperiness/dishoniesty tends to be characterized by wordsreferring to fish, which are slippery and elusive; slyness/evasiveness tends to becompared to the nature of the fox--a sly and elusive predator. Wolves, tigers,lions, cats, and dogs all have contributed to modern idiom formation acrosslanguages and cultures. So general is this tendency to build from already existingmaterials that entire proverb-length statements also can be found in differentlanguages and cultures. Consider English "don't count your chickens beforethey're hatched," which means something such as "refrain from celebrating thesuccess of an undertaking prematurely lest it misfire involving you in embarrass-ment." In the unrelated Hungarian, one says "Ne igydl elbre a medve bbrjre,"which translates as "do not drink [a toast] to the bear's hide in advance." Thesame idea is expressed in French as "Ne vendez pas la peau detours avant del'avoir tug," echoed by the Italian "Non vendere la pelle d'oso prima di averlopreso" ("do not sell the hide of the bear before having taken [captured] it").Russian has "Medved' es6e v lov'e, a ty uie medved'a porodadsi'" ("the bear isstill in the hunt [being hunted], but you are already selling it [the bear]"). It isfairly transparent, I think, that these five different versions in five differentlanguages, by using different words and making different propositions still aresaying, at a deeper level, the same thing. Certainly, any translator of a literary orcolloquial text who found this saying in any of the language would be welladvised to try to find the corresponding idiom in the target language and to use itin its totality instead of laboriously translating the idiom word for word.

This fact of the popularity of animals is interesting because the English

13. IDIOMATICITY AND L'ARBITRAIREDU SIGNE

version of "Don't count your chickens . . ." can be traced back to the fables ofAesop, whereas the continental versions involve a hunting scene with a resultingcelebration or the anticipated sale of the hide of a bear. The common underlyingmoral, that the premature celebration of success can end in failure and embarrass-ment, must have been a common experience to the societies that have thisproverb. The versions using the "bear" as the object of capture may have beeninfluenced by one another-certainly the French and the Italian are much tooclose to ignore the similarity-the Hungarian mentions "drinking"; the Russian,as well as the French and the Italian, are concerned with the "sale" of the hide. Itseems that comparative literature and ethnographic history have much unfinishedbusiness in common in digging up the exact route such proverbs have traveledover the centuries.

The tendency toward internal borrowing of materials created via phonopoeticarbitrariness into the higher level of imagopoeia is at the heart of idiom formationon a global basis. In my article "Idiomaticity as a Language Universal" (Mak-kai, 1978), I gave a long list of heavily reinvested words such as French coup("strike" or "blow"), which has produced hundreds of idioms. In this chapter Ihave suggested that idiom formation is a reaction to the arbitrariness of earliermodes of sign formation, chiefly by the less conscious phonopoetic approach tologopoeia. Figure 13.7 (overleaf) presents this hypothesis of logopoetic evolu-tion from phonopoeia to imagopoeia. I regret that I cannot at this stage of mywork on idioms offer irrefutable proof that languages always evolve in thisfashion. It is, of course, possible that a limited number of complex expressionsare image based from the very start just as it is possible, though a rather rareevent, for modern languages to introduce onomatopoetic or other pho-nopoetically coined lexical items that then are accepted socially.

It was the Italian jurist, historian, and philosopher, Gianbattista Vico(1770/1961) who in his recently rediscovered and increasingly influential bookon science first suggested in Western speculation on language that the mostancient form of communication was probably a gestural one. 1 have devoted alonger article to Vico's relevance in modem linguistic speculation (Makkai, inpress b), which will be included in my collection of essays (Makkai, in press a).

In closing then, we should observe that idioms are the price we pay for thevast proliferation of concepts vis-A-vis the relatively arrested development of ourphonological abilities. Here, again, linguistic ontogeny may be seen to recapitu-late linguistic philogeny: Children have a much more flexible phonetic-artic-ulatory apparatus than even young adults after the age of 21 years. Children canlearn a foreign language without a "foreign accent" without any effort-after theage of 21 years most people speak a foreign language with an appreciable foreignaccent. This foreign accent, which shows their phonetic limitation, however,does not preclude people older than 21 years to acquire new concepts, indeed toacquire a college or graduate education in many a technically exacting field.Conceptual development thus easily continues when phonetic (hence pho-

321

322 MAKKAI

1. TIE FIRST PH5SE OF LOGOPOEIAvia gesturing or XIETOPOEIA;gestural c. 1icatioc my betermed 'kinetopoetic. (A)

THE TWO ANCIENT PHASES OF NON-ARBITRARINESS

It. THE SECOND PHISE OF LOCOPOCIAvia soumd-iitation orOMTOPOEIA; soemd-iiitatinmg

woa gestre say be termed'onomatopoetic' comunication.

-1 .111. THE THIID PHSE OF LOGOPOEIA

via the increasingly arbitraryallocation of available sound-strigs to act as carriers ofeesn1 g.\Expd1ing vocabulary sets ofmay thousand istms rum intoPHONOPOETC SATURATION as soowas HQWWItlY asd POLYSENtY ke \the system auard to handle oathe spoken level (Chiese. English). \(B)

1Wt IH U K MU ULK rFIUEI yrZ Ur miAL. KIKSNtfl

,AND NON-ARBITRARINESS RESULTING IN 'DOUBLE VISION'IV. TIE FOIM PMaE OF LOGOPOEIA / FOR MOST CONTEMPORARY IDIOMtS

via image-based metonymic extensioeof coound nominal verbal and /adjectival fors snd their ifstitu-/tionalliatiom in societies without/regard to their 'literal' origins.Characterized by heavy vocabularyexpansion and the gradual predominanceof wri tten records.

FIG. 13.7. A schematic outline of human semeio-genesis in natural

languages.

nopoetic) flexibility is no longer there. Idioms almost always can be literalized,but such literalizations are perceived either as comical, exaggerated, or based onhistorical hindsight, which is not immediately relevant to a synchronic descrip-tion of the use of the form at hand.

REFERENCES

Algeo, J. (1970). English: An introduction to language. New York: Harcourt, Brace & World.

Bloomfield, L. (1933). Language. New York: Holt.

Bolinger, D. L. (1976). Memory and meaning. Forum Linguisticum, 1, 1-14.

Bolinger, D. L. (1977). Idioms have relations. Forum Linguisticum, 2, 157-169.

Chisolm, L. W. (1963). Fenollosa: The Far East and American culture. New Haven, CT: Yale

University Press.

Chomsky, N. (1965). Aspects of the theory of synIax. Cambridge, MA: MIT Press.


De Saussure, F. (1959). A course in general linguistics (W. Baskin, Trans.). New York: PhilosophicalLibrary. (Original work published 1915)

Eliot, T. S. (1963). Literary essays of Ezra Pound. London: Faber & Faber.Fabricius-Kovdcs, F. (1961). A propos d'une loi semantique [Regarding a semantic law]. Acta

Linguistica Hungarica, 11, 405-411.Fenollosa, E. F. (1920). The Chinese written character as a medium for poetry. In E. Pound,

Instigations of Ezra Pound together with an eassay on the Chinese written character by ErnestFenollosa. New York: Boni & Liveright.

Flavell, R. H. (1972). Dictionnaire des locutions idiomatiques anglaises employant pertains motsde couleur [Dictionary of English idiomatic expressions using certain color words]. Unpublisheddoctoral dissertation, Universitd de Paris.

Halliday, M. A. K. (1975a). Language as social semiotic. London: Edward Arnold.Halliday, M. A. K. (1975b). Learning how to mean. London: Edward Arnold.Hockett, C. F. (1958). A course in modern linguistics. New York: MacMillan.Koestler, A. (1967). The ghost in the machine: The urge to self-destruction: A psychological and

evolutionary study of modern man's predicament. New York: Macmillan.Kronasser, H. (1952). Handbuch der semasiologie [Handbook of semology]. Heidelberg: Karl

Winter Universitatsverlag.Labov, W. (1964). The social stratification of English in New York City. Washington, DC: Center

for Applied Linguistics.Labov, W. (1976). On the use of the present to explain the past. In A. Makkai, V. B. Makkai, & L.

Heilmann (Eds.), Linguistics at the crossroads (pp. 226-261). Padova and Lake Bluff, IL:Liviana Editrice and Jupiter Press.

Lamb, S. M. (1966). Outline of stratificational grammar. Washington, DC: Georgetown UniversityPress.

Lamb, S. M. (1990). Linguistic model and linguistic thought: The case of "either-or" thinking. InThe 17th LACUSforum (pp. 109-120).Lake Bluff, IL: LACUS, Inc.

Makkai, A. (1972). Idiom structure in English. The Hague, Netherlands: Mouton.Makkai, A. (1976). Idioms, psychology, and lexemic principle. In The 3rd LACUSforum (pp. 467-

478). Columbia, SC: Hornbeam.Makkai, A. (1978). Idiomaticity as a language universal. In Universals of human language: Vol. 3.

Word structure (pp. 401-448). Stanford, CA: Stanford University Press.Makkai, A. (in press a). Ecolinguistics: Toward a new paradigm for the science of language.

Budapest and London: Akademiai kiad6 and Frances Pinter.Makkai, A. (in press b). On positive and negative logic in modem linguistic theorizing: A Vichian

perspective on the development of the philosophy of language and semiotics. In M. Danesi (Ed.),Proceedings of the 1990 University of Toronto symposium on semiotics. Berlin: Mouton-De-Gruyter.

Makkai, A., Makkai, V. B., & Heilmann, L. (Eds.) (1976). Linguistics at the crossroads. Padovaand Lake Bluff: Liviana Editrice & Jupiter Press.

Pound, E. (1920). Instigations of Ezra Pound together with an essay on the Chinese written char-acter by Ernest Fenellosa. New York: Boni & Liveright.

Pound, E. (1963). Literary essays of Ezra Pound. London: Faber & Faber.Rohsenow, J. S. (1984). Idiomaticity in Mandarin Chinese. In The 11th LACUSforum (pp. 323-

340). Columbia, SC: Hornbeam.Ruhl, C. E. (1974). Primary verbs. In The Ist LACUS form (pp. 436-446). Columbia, SC:

Hornbeam.Ruhl, C. E. (1975). Pragmatic metonymy. in The 2nd LACUSforum (pp. 370-380). Columbia, SC:

Hombeam.Ruhl, C. E. (1976). Idioms and data. In The 3rd LACUS forum (pp. 456-566). Columbia, SC:

Hornbeam.

324 MAKKAI

Ruhl, C. E. (1978). Alleged idioms with "hit." In The 5th LACUSform (pp. 93-107). Columbia,SC: Hombeam.

Ruhl, C. E. (1979). The semantic field of "break," "cut," and "tear." In the 6th LACUSforum(pp. 200-214). Columbia, SC: Hombeam.

Ruhl, C. E. (1980). The noun "ice." In The 7th LACUS forum (pp. 257-269). Columbia, SC:Hornbeam.

Ruhl, C. E. (1982). Figurative "kick." In The 9th LACUSforum (pp. 301-310). Columbia, SC:Hornbeam.

Ruhl, C. E. (1989). On metonymy. Buffalo: SUNY at Buffalo Press.Vicb, G. (1770/1961). La szienza nuova [The new science] (M. Fisch & E. Bar, Trans.). Ithaca,

NY: Cornell University Press.Welch, D. H. (1974). Latinate and English verbs and nouns: A synchronic, diachronic and pan-

chronic description. In The ]st LACUSforum (pp. 458-471). Columbia, SC: Hombeam.Yip, Wai-lim (1969). Ezra Pound's CATHAY. Princeton, NJ: Princeton University Press.Yngve, V. (1974). The dilemma of contemporary linguistics. In The Ist LACUSforum (pp. 1-16).

Columbia, SC: Hombeam.Yngve, V. (1986). To be a scientist. In The 13th L4CUSforum (pp. 5-25). Lake Bluff, IL: LACUS,

Inc.

Author Index

Note: Italicized page numbers refer to bibliography pages.

A

Abeille, A., 251, 256, 270Ackerman, B. P., 43, 50, 105, 112, 119,

126Algeo, J., 300, 322Alshawi, H., 265, 270Alston, W., 59, 75Altmann, G. T., 157, 206, 223Anderson, J., 187Anderson, S. R., 51, 158, 246Antos, S. J., 5, 26, 62, 77, 134, 141, 146,

158, 165, 186, 204, 224, 229, 246Arens, Y., 237, 247Attneave, F., 156, 157

B

Babby, L., 289, 294Baddeley, A., 50Baker, M. C., 292, 294Balota, D. A., 171, 186Bard, E. G., 154, 157Barlow, J., 103, 128Beeman, M., 217, 223Bell, S., 4, 5, 25, 57, 75, 84, 97, 148, 157,

164, 186, 273, 294Berndt, R. S., 217, 223Bienkowski, M., 163, 187Bihrle, A. M., 217, 219, 223

Bloomfield, L., 317, 322Boatner, M., 72, 75, 276, 279, 291, 294Bobrow, S., 4, 5, 25, 57, 75, 84, 97, 148,

157, 164, 273, 294Boisset, J. H., 130, 140Bolinger, D. L., 315, 322Bolton, J. L., 6, 26, 32, 34, 51, 63, 76, 102,

127, 234, 246, 272, 294Bookin, M. B., 43, 51Bortolini, U., 167, 186Botelho da Silva, 138-139, 141Brannon, L. L., 135, 141Bresnan, J., 251, 253, 255, 261, 270Brito, A. M., 132, 141Broderick, J. P., 38, 50Brown, R., 18, 26Brownell, H. H., 217, 219, 223Brugman, C., 65, 76,Burgess, C., 209, 216, 219, 223-224Bums, B. C., 27, 52

C

Cacciari, C., 6, 9, 13-14, 17-18, 26, 34, 37-38, 43, 49, 50-52, 65, 76, 83-84, 90,97-98, 103, 105, 119, 123, 125, 126,134, 141, 145, 148, 150, 155, 157, 159,165-166, 171, 173, 183, 186-187, 206,221, 223, 231, 234, 238, 243-244, 245,276, 287, 294-295

325

326 AUTHOR INDEX

Canter, G. J., 134, 141, 146-147Caplan, D., 217, 223Caramazza, A., 217, 219, 223Carlson, M., 85, 98, 207, 224Carlson, T. B., 27, 50Carpenter, P. A., 222, 224Chafe, W., 34, 50Charles, W. G., 283, 294Chiarello, C., 216, 219, 223-224Chisolm, L. W., 305, 322Chomsky, N., 156, 157, 166, 186, 251-252.

254-255, 261, 270, 277, 294, 316, 322Clark, E., 28, 50Clark, H. H., 28, 31, 43, 50, 229, 245, 294Clifton, C., 85, 97, 207, 216, 224Cohen, J. D., 184, 186Cole, P., 52, 77, 295Colombo, L., 163, 170-171, 186-187Coltheart, M., 98Cottrell, G. W., 76, 223, 237, 246Coulmas, F., 27, 37, 42, 50Crain, S., 205-206, 224Cronk, B., 48, 50, 148, 156, 158Cruse, D. A., 19, 26, 31, 33, 34, 50, 291,

294Cutler, A., 4, 5, 12, 26, 33, 52, 57, 63-64,

76-77, 84, 98, 101, 128, 134, 136, 140,141, 146-147, 159, 164-165, 187, 203--204, 208, 224, 229, 233, 238, 246, 273,287, 294-295

Cutting, C., 6, 7, 13, 16, 17, 20, 26, 32, 35.51, 63-64, 66, 76, 125, 127, 173, 186,234, 246

D

Danesi, M., 323Dascal, M., 27, 28, 29, 50, 101, 127, 145,

158Davies, M., 34, 50De Saussure, F., 297, 323Dell, G. S., 209, 224Dowty, D., 65, 76, 224Duffy, S. A., 164, 186-187Dunbar, K., 184, 186Dyer, M., 237, 246

E

Eberhard, K., 209, 224Eliot, T. S., 305, 323

Ellis, A. W., 225Elman, J. L., 147, 158Engel, M., 112, 128Estill, R. B., 57, 76, 134, 141, 147, 158,

204, 224

F

Fabricius-Kovacs, F., 311, 323Farkas, D., 50, 52Fellbaum, C., 283, 294Fenollosa, E. F., 305, 323Fernando, C., 31, 32, 50, 145, 158Ferreira, F., 85, 97, 207, 224Fillmore, C. J., 27, 29, 41, 50, 62, 76, 288,

294Fine, H., 103, 128Fischer, U., 283, 294Flavell, R. H., 31, 32, 50, 314, 323Flores d'Arcais, G. B., 50, 85, 97, 141Fodor, J., 27, 51, 265, 270Foldi, N. S., 217, 224Forster, K. 1., 171, 186Fox, B., 292, 294Fraser, B., 33, 51, 80, 98, 106, 127, 138,

141, 145, 158, 164, 166, 186, 208, 224,251, 270, 272, 287, 294

Frauenfelder, U., 158Frazier, L., 85, 98, 164, 172, 174, 183, 185,

186-187, 207, 216, 224Freedle, R., 294Fromkin, V. A., 14, 26, 38, 51

G

Gaatone, D., 34, 51Gardner, H., 112, 128, 217, 219, 223, 225Garrett, M., 159, 209, 225Gates, J., 72, 75, 276, 294Gazdar, G., 173, 183, 186, 251, 255, 257-

261, 268, 270Gazzaniga, M. S., 217, 224Gerrig, R. J., 43, 50-51Gibbs, R. W., 4-7, 12-13, 16-17, 20, 23-

24, 26, 27-29, 32, 34-37, 42-43, 51-52, 57, 61-64, 66-67, 70, 72, 76-77,84, 98, 101-102, 112, 119, 124-125,127, 134, 136, 138, 140, 141, 145-148,158, 164-165, 171, 173, 183-185, 186,204-206, 208-209, 212, 224, 229, 231-234, 246, 272, 276-278, 286-287, 294

AUTHOR INDEX 327

Gildea, P., 43, 51Glass, A. L. 147, 158, 165, 186Glucksberg, S., 6, 9, 13-14, 17-18, 26, 27-

29, 31, 34, 37-38, 43, 50-51, 83, 97,123, 125, 127, 145, 157, 173, 183, 186,231, 233-234, 245-246, 276-278, 284,294

Gonzales, G. P., 35, 51, 61, 64, 76, 102,127, 134, 136, 138, 140, 141, 164-165,184, 186, 208-209, 212, 224

Goodman, G. O., 209, 224Graesser, A. C., 37, 51Greciano, G., 130, 141Green, G., 58, 76Greenbaum, S., 281, 295Greenberg-Concool, N., 9, 26Gross, D., 283, 294

H

Halle, M., 51Halliday, M. A. K., 298, 301, 323Hamby, S., 217, 225Hattori, S., 159, 295Haviland, S., 284, 294Heeschen, C., 217, 224Heilmann, L., 323Hendrix, G. G., 237, 246Heringer, J., 61, 76Hildebrandt, N., 217, 223Hillyard, S. A., 217, 224Hirst, W., 217, 224Hobbs, J., 32, 51Hockett, C. F., 27, 31, 51, 307, 312, 323Hoffman, R. R., 27, 51Holland, D., 36, 51Honeck, R. P., 27, 51, 112, 127Hub Faria, 1., 132, 141

Inoue, K., 159, 295

J

Jacobs, P., 237, 246Jacobsen, W. M., 50, 52Jarvella, R. J., 50Job, R., 163, 187

Johnson, M., 24, 26, 60, 63, 66, 76Jurgens, R., 217, 224Justeson, J. S., 283, 294

K

Kaplan, R., 239, 246Kartunnen, L., 224Katz, J. J., 27, 33, 51, 106, 127, 145, 158,

238, 246, 251, 283, 294Kay, P,. 41, 51, 62, 76, 239, 246, 288Kemper, S., 57, 76, 105, 113, 127, 134, 141,

147, 158, 204, 224Kempler, D., 217-220, 225Keppel, M. E., 6, 26, 32, 34, 51, 63, 76,

102, 127, 234, 246, 272, 294Keysar, B., 18, 26Kiparsky, P., 51, 158, 246Klein, E., 173, 186, 251, 270Koestler, A., 298, 305, 323Koriat, A., 172, 186Kdvecses, Z., 70, 76Kraaij, W., 237, 246Kreutz, R. J., 37, 51Kronasser, H., 311, 323Kronfeld, X., 59, 76Kutas, M., 163, 172

L

Labov, W., 316, 323Lahiri, A., 154, 158Lakoff, G., 24, 26, 29, 30, 32, 36, 37, 51,

60, 62-63, 65-67, 76, 146, 158, 231,246

Lamb, S. M., 301, 309, 316-317Langacker, R., 62, 76Langer, P., 171, 187Laver, J., 187Le Doux, J., 217, 224Leach, E., 119, 127Lehnert, W. G., 34, 52Lehrer, A., 27, 31, 52Leiman, J. M., 163, 187Levelt, W. J. M., 50, 97Levi, J. N., 270Levinson, S. C., 33, 52Levorato, M. C., 43, 50, 52, 103, 105-106,

119, 127Lewis, D., 58, 76Linebaugh, C. W., 217, 224

328 AUTHOR INDEX

Lodge, L., 119, 127Long, J., 50Long, T., 72, 76Lorch, R. F. J., 171, 186Luce, P., 154, 158Lucy, P., 229, 245Lupker, S. J., 170-171, 186

M

Makkai, A., 31, 52, 72, 75, 130, 141, 276.302, 311, 321, 323

Makkai, V. B., 323Markman, E., 119, 127Marslen-Wilson W. D., 147, 154, 158, 205,

224McClelland, J. L., 147, 158, 184, 187, 206.

209, 224McDonald, J. L., 222, 224McGlone, M. S., 6, 9, 11, 37, 51, 127, 276,

294McRae, K., 48, 52, 156, 158Medin, D., 43, 52Mehler, J., 159Michelow, D., 217, 223Michiels, A., 27, 42, 52Miller, G. A., 283, 294Mira Mateus, M. H., 132, 141Mistler-Lachman, J. L., 106, 127Montague, R., 257, 270Morgan, J. L., 52, 58, 77Morris, R. K., 164, 187Mueller, R. A. G., 42, 52, 141, 147, 158Murphy, G. L., 43, 51, 52Myers, P. S., 217, 224Myers, T., 187

N

Nagy, W., 30, 52Nayak, N. P., 6-7, 13, 16, 17, 20, 24, 26,

27, 32, 34-37, 42, 51-52, 62-64, 66,70, 76-77, 102, 125, 127, 134, 141,145, 158, 173, 183, 186, 231-234, 246,272, 294

Nelson, K. E., 120, 127Newmeyer, F., 251, 268-269, 270, 272, 280,

287, 295Nippold, M. A., 105, 119, 127Noteboom, S. D., 154, 158

Nunberg, G., 5, 8, 13, 17, 26, 27, 28, 30,34, 35, 42, 52, 62, 77,173, 187, 234,246, 251, 270, 271-273, 279-280, 286,289, 292, 295

0

O'Brien, J. E., 36, 51, 63, 67, 72, 76, 145-146, 158-159, 231-232, 246

O'Connor, M. C., 51, 62, 76, 288Onifer, W., 163, 183, 186Ortony, A., 5, 6, 26, 52, 62, 64, 77, 112,

128, 134, 141, 146, 158, 165, 186, 204,221, 224, 229, 246

Osherson, D., 119, 127

P

Palmer, F., 71, 77Parret, H., 50Pergateau, P. M., 270Peters, S., 65, 76Peterson, R. R., 209, 224Phares, C., 270Piaget, J., 119, 125, 127-128Pisoni, D. B., 154, 158Pollio, H. R., 27, 52, 103, 128Polio, H., 103, 128Pollock, A., 216, 224Popiel, S. J., 48, 52, 156, 158Postal, P. M., 106, 127Potter, H. H., 219, 223Pound, E., 305, 323Powelson, J. A., 217, 223Prince, E., 284, 295Prinz, P. M., 112, 119, 128Puhvel, J., 77, 98, 141Pullum, G. K., 173, 186, 251, 270Pulman, S. G., 265, 270, 271, 284, 295

Q

Quinn, N., 36, 51Quirk, R., 281, 295

R

Randall, J., 224

AUTHOR INDEX 329

Rayner, K., 85, 98, 164, 172, 174, 183, 185,187, 207, 224

Reagan, R. T., 37, 39, 135-136, 141, 208,224

Reynolds, R. E., 5, 26, 62, 77, 134, 141,146, 158, 165, 186, 204, 224, 229,246

Richards, L, 216, 224Rohsenow, J. S., 309, 323Rosen, A., 216, 223Ruhl, C. E. 279, 295, 309, 314-315, 323-

324Rumelhart, D. E., 31, 52, 146, 158Ruwet, N., 32, 34, 35, 37, 52, 272, 295

S

Sag, I., 8, 26, 34, 52, 145, 159, 173, 186-187, 234, 246, 251, 270, 271, 295

Salasoo, A., 154, 158Sanders, M., 171, 187SisbA, M., 50Schallert, D. L., 5, 26, 62, 77, 134, 141,

146, 158, 165, 186, 204, 224, 229,246

Schweigert, W. A., 39, 52, 84, 98, 113, 128,156, 158

Searle, J., 27, 43, 52Seidenberg, M. S., 163, 171-172, 187Shannon, C. E., 156, 158Shibtani, M., 76Shillcock, R. C., 154, 157Shoben, E., 43, 52Silva Duarte, I., 132, 141Simpson, G. B., 26, 50, 97, 127, 157, 163,

186-187, 216, 219, 223, 245Small, S. L., 76, 223, 246Sowry, B., 112, 127Stanovich, K. E., 209, 225Steedman, M., 205-206, 223-224Stein, S., 217, 224Stock O., 238, 240, 246Stroop, J. R., 6, 26Summers, D., 72, 76Svartvik, J., 281, 295Sweetser, E., 59, 77Swinney, D. A., 4-5, 12, 26, 33, 52, 57, 64,

77, 84, 98, 102, 128, 134, 141, 146-147, 154, 158-159, 163-165, 183, 186-187, 203-204, 208, 224, 229, 233, 238,246, 273, 295

T

Tabossi, P., 6, 12, 26, 49, 50, 52, 65, 76, 84,90, 97-98, 125, 127, 134, 141, 146-148, 150, 154-155, 157, 159, 163, 165-166, 171, 183, 186-187, 206, 221, 223,244, 245, 287, 295

Tagliavini, C., 167, 186Talland, G. A., 294Tanenhaus, M. K., 76, 163, 187, 223, 246Tannen, D., 33, 52Taraban, R., 206, 224Tarrant Martin, S., 105, 119Todrys, K. W., 50, 52Turner, M., 30, 32, 51, 60, 76Tyler, L. K., 158, 205

U

Ullmann, S., 36, 52

V

Van der Linden, E., 237-238, 244, 246Van Lancker, D., 134, 141, 216-220, 225Van Petten, C., 163, 172, 187Vico, G., 321, 324Voegtle, K., 112, 127Vosniadou, S., 112, 128

W

Walker, E. C. T., 159Wall, R., 65, 76Waltz, D., 237, 246Wapner, W., 217, 225Wasow, T., 8, 20, 26, 34, 52, 145, 159, 173,

183, 187, 234, 246, 251, 257, 270, 271-272, 274, 279, 280, 291, 295

Waters, G. S., 171, 187Waugh, C., 294Weaver, W., 156, 158Weinreich, U., 57, 77, 83, 98, 130, 141Welch, D. H., 316, 324West, R. F., 209, 225Wierzbicka, A., 287, 295Wilensky, R., 237, 247Williams, J., 170-171, 186Winner, E., 43, 52, 103, 112, 128

330 AUTHOR INDEX

Wood, M. W., 230-231, 247Wright, B., 209, 225

Y

Yip, Wai-lim, 305, 324Yngve, V., 316, 324

z

Zampolli, A., 167, 186Zerik, U., 237, 246Zurif, E. B., 217, 219, 223Zwicky, A. M., 224Zwitserlood, P., 154, 159

Subject Index

A

Acceptability, 136-139, 278Access, 134

direct, 4, 5, 9, 12, 23, 164, 203exhaustive, 163, 165idiom, 5, 84, 165, 244

meaning, 5lexical, 5reordered, 164selective, 163- 164

Acquisition, 43idiom, 37, 112-113, 115, 119, 122, 123,

125idiomatic meaning, 126language, 44, 101 -103, 121, 123, 298

Activation, 184idiom, 150, 154, 164-165, 178, 243meaning, 65, 129, 147-149, 152-153, 155,

163-167, 170-175, 177, 181, 185,244

semantic, 147-148, 153, 155, 164, 172word, 171, 244

Ambiguity, 81, 106, 115, 130, 146, 157, 164-167, 174-175, 177, 182-185, 216,234, 237

lexical, 163, 237literal vs. figurative, 146

Analogy, 95, 120

Analysis,compositional, 58, 64-66, 203, 204linguistic, 4, 5, 6, 9, 11, 17, 22-23semantic, 19, 87, 106, 110, 203, 215, 218,

222-223syntactic, 19, 81-82, 85, 87, 89-91, 97,

106, 203, 208, 210, 212-215, 217-218, 222-223, 257

Arbitraire, see ArbitrarinessArbitrariness, 36, 58, 62, 67, 71, 74, 121,

123, 232, 303, 318-319, 321du signe, 297-298, 319

C

Communicative,intent, see Intentintention, see Intentionneed, 38

Competence, 102, 110, 119, 123, 277-278,316

figurative, 43, 103-104, 106, 110, 112-113, 115-117, 119-123

idiom, 44grammatical, 274lexical, 105metalinguistic, 119morphosyntactic, 105

331

332 SUBJECT INDEX

Compositional, 4, 5analysis, see Analysissemantics, 35, 249, 251, 261syntax, 251

Compositionality, 5, 13-14, 16, 19, 22, 33-34, 48, 79, 82, 100, 145, 231, 249,272

Comprehension, 21, 25, 28-30, 75, 80, 86,89, 109, 117-118, 120, 124, 149, 169,172, 176, 202, 205, 207, 209, 212-213, 217, 219-220, 229, 244, 317

figurative language, 29, 57, 112figurative meaning, 105, 108, 165idiom (or idiomatic), 4, 5, 6, 7, 8, 11, 19,

25, 32, 48, 57-58, 64-65, 74, 79, 83-84, 86, 89, 91, 96, 98, 104, 108, 113,116, 122-125, 129, 134, 145-146,148, 156, 163-164, 183, 185, 202-203, 208-210, 216, 218, 219, 222,229, 231, 233-235, 237-238, 243-244, 276, 277

language, 84, 85, 103-104, 157, 201, 216literal language, 29, 57, 147literal meaning, 165metaphor, 61, 83

Computation, 93, 172, 177, 201, 240idiom, 97, 202, 220meaning, 42, 84, 86, 146, 165, 203-204,

206-208, 214, 221Concept, 18, 122, 183, 232-233, 275-277,

283, 297, 299, 302, 306, 309, 319complex, 24, 43, 232

Conceptual,mapping, see Mappingmetaphor, see Metaphorreferent, see Referent

Constraint, 318pragmatic, 20, 41-42semantic, 7, 20, 30, 41-42syntactic, 23, 239, 244

Content,allusional, 5, 18, 23-24

Context, 5, 22, 37-38, 42-43, 57, 71, 73-74,81, 84, 87, 91-93, 95, 102-103, 105-109, 112-115, 117, 120-122, 124-126, 137, 147-148, 155-156, 163-166, 170-172, 174-176, 181-183,202, 204-205, 209, 211-212, 214,217, 221, 222, 232, 239, 240, 245,249-252, 262, 264-265, 268-269,275, 278-284, 288, 303, 305

biasing, 93, 148, 163-164, 166-167, 172-175, 177-178, 181-183, 185, 192,195, 204-206, 208, 212-215, 217,220-222

discourse, 4, 23, 24figurative, 168, 180, 187, 195, 210idiom, 10, 13idiomatic, 91-92, 109-110, 113-114, 118,

120, 169-170, 177, 179, 180, 182,209-210, 275

literal, 91-92, 109, 113-115, 118, 168-171, 177, 179, 181-182, 204, 209-210, 214

Counterpart, 129, 130-131, 135, 140literal, 213, 258, 279

D

Determiner, 272-275, 278, 280, 282, 285,291-293

definite, 280, 282, 284, 286-287, 293indefinite, 272, 291, 293negative, 273, 286

Discourse, 16productivity, see Productivityrule, see Rule

Dissociation, 203, 216

E

Expression, 4idiomatic, 5linguistic, 5, 6literal, 5, 9, 10, 11, 12

F

Familiarity, 66, 74, 81, 83, 86-97, 113-118,122, 156-157, 169

Flexibility,idiom, 14-15, 20, 80, 124-125, 208, 241,

272, 280, 291lexical, 8, 15-16, 63semantic, 126, 235syntactic, 16, 208, 213, 234-238, 245, 278

Formcanonical, 254-257, 261, 276-278idiom, 10

SUBJECT INDEX 333

lexical, 10logical, 257, 262-265, 267-268semantic, 238syntactic, 10, 35, 129, 139, 184, 238, 254,

256, 263, 268Formedness, 277

ill-, 65-66, 129-132, 134-137, 139-140well-, 65-66, 79, 135-137, 139-140

G

Grammar,Generalized Phrase Structure (GPSG), 251,

259Lexical-Functional, 251Montague, 257Standard Theory transformational, 251transformational generative, 316Tree Adjoining (TAG), 251, 256-257

Grammaticaltheory, see Theory

H

Hemispheric,lateralization, 203, 216

Identification, 85word, 155, 172

Idiom (or Idiomatic), 3, 4, 5, 6, 7, 9, 10, 24,27, 29-34, 40, 42, 49, 57-59, 61-62,65, 164, 272, 276

acquisition, see Acquisitionanalizibility, 27, 32, 35, 39, 62-64, 231,

234-236, 242compositional, 5, 6, 13, 15, 17, 19-20, 22-

23, 35, 74, 271-272, 278, 283, 293constituent, 8, 10, 14, 19-20, 35, 37, 231,

234-235, 237, 242-243, 256, 271,273, 277

decomposition, 17, 35, 37, 62, 65, 79, 173,271

development, see also Acquisitionfamiliar, 5, 8, 9, 10, 11, 39, 41flexibility, see Flexibilityformation, 57, 63, 80-81, 84, 105, 135-

137, 139, 145-146, 164-165, 184,309, 318, 320-321

frozenness, 16, 30, 57, 63, 80-81, 84, 105,135-137, 139, 145-146, 164-165,184, 208, 212-213, 219, 272, 278,286, 292, 293

key, 147-149, 151-156, 287list, 4, 5, 47, 83, 164mental, 4motivation, 35noncompositional, 6, 15-17, 19-20, 23,

34, 61, 271-272, 278, 283, 293opaqueness, 15, 17, 20-21, 35, 37-39, 43-

44, 46-47, 65, 80-81, 83, 87, 91-95,145, 185, 231, 235, 271, 286-287,291

phrasal, 4, 7, 15, 24, 35, 79, 82, 89quasi-metaphorical, 22, 24, 45structure, see Structuretransparency, 15, 17, 20-21, 37, 39, 43-

44, 46, 49, 80, 88, 91-92, 95, 145,181, 185, 231, 234, 271, 287, 291,308-309

transformability, 135- 137, 139-140variant, 8, 9, 10, 11, 12, 14-15, 277variability, 14, 19, 250, 275

Idiomaticity, 27-28, 31-33, 57, 59, 62, 69,71, 74-75, 93, 129-130, 134, 136,145, 290, 303-305, 309, 314, 318

Image, 41, 320mental, 36, 67-69

Imagery, see ImageInference, 37, 42, 72, 102, 104-105, 111,

122, 165, 217, 262-266, 268, 283quasi-, 261, 263-264

Intent,communicative, 8, 14-15, 20-23

Intention,communicative, 101, 104, 121-122, 125

Interpretation, 17, 22-25, 29, 32, 36-37, 39-41, 45, 49, 64, 70, 74, 79, 81, 84-85,87-88, 91, 93-97, 105, 115, 121,123, 145, 165, 172-174, 178, 181-185, 201-202, 204-205, 215, 219-220, 222, 237-241, 243-244, 250-251, 254, 257, 260, 263-266, 272,281-282, 284, 288-290, 292-293

figurative, 62-66, 71, 73, 108, 115, 146,148, 166, 170, 174-175, 177, 179,181 -183, 187, 202, 204-206, 208,214, 217, 219-222, 245, 271, 275

334 SUBJECT INDEX

Interpretation, (cont.)idiomatic, 39, 81, 91, 96, 105, 114, 134,

171-172, 176-177, 183, 185, 217-218, 221, 243, 245, 262-264, 266,269

literal, 48, 66, 80-82, 84-86, 88-89, 93,108-109, 114-115, 120, 125, 129,134, 146, 148, 163-166, 171-172,174-175, 177, 183, 185, 187, 202,204-206, 214-215, 217, 221-222,229, 243, 245, 252, 262-264, 271,275-276, 278, 280-281, 283-284,288, 292

non-literal, 64-65, 68metaphorical, 278pragmatic, 4semantic, 28, 33, 93-94, 115, 120, 145,

205, 239, 245, 257Irony, 28, 103-104

K

Knowledge, 45conceptual, 24, 232linguistic, 238, 274world, 14, 28

L

Languagefigurative, 25, 31, 42-44, 48, 57, 75, 83,

101-104, 110-111, 121-123, 146,217, 272, 274, 279

literal, 28-32, 34, 42, 216, 272-276, 278-281, 284, 286-289, 291-293

non-literal, 271Lexical,

access, see Accessambiguity, see Ambiguitydecision, 147-151, 154, 166-169, 171, 209flexibility, see Flexibilityitem, 31, 42, 70, 85, 140, 146, IS10-51,

154-155, 203, 218-219, 230, 233,250, 254-255, 259, 265-267, 271-273

memory, see Memoryprocessing, see Processingoperations, 6rule, see Rulerepresentation, see Representation

substitution, 7, 14-15, 19-21, 122, 272,276, 286

unit, see Unitvariation, 122

Lexicon, 49idiom, 4, 10mental, 5, 10, 41, 57, 203, 232, 273, 275,

283word, 4

Linguistic,processing, see Processing

Literal,interpretation, see Interpretationmeaning, see Meaningprocessing, see Processing

Literality, 30-31Logic,

intensional, 257, 260-261sorted, 259

Look up, 4, 250direct, 4, 6idiom, 6, 11idiom-meaning, 6vs compositionality, 6vs direct access, 4

M

Mapping,conceptual, 59, 67, 70metaphorical, 35, 39, 59, 61, 63, 66, 69-

71, 74referent, 235-237

Meaning, 3, 5, 6, 8, 9, 11, 12, 23, 25, 28,31, 35, 37-38, 58, 61, 64, 66-67, 71,75, 79, 82, 85, 89, t02, 104-107,III, 113, 119-126, 129, 134, 145,147, 151, 154-157, 163-165, 174,184, 204, 208, 212, 217, 220, 231,238, 249-250, 253, 257, 275-276,278, 301, 304-305, 316

allusional, 23-25compositional, 66, 163, 166, 251, 263computed, 36conventional, 15, 29, 39, 96, 97, 102expression, 4figurative, 29, 33, 35-36, 38-39, 43, 48,

57, 61-71, 73-74, 94, 102, 104-105,107-108, 113, 115, 122, 125, 146,

SUBJECT INDEX 335

157, 163-168, 170-175, 177, 180-183, 185, 192, 195, 203-206, 208-209, 214-215, 218, 221-223, 237,272-274, 276, 278-279, 284, 286-287, 293

idiom, 4, 5, 6, 8, 9, 13-14, 17, 19-22, 40,42, 48-49

idiomatic, 3, 4, 5, 6, 7, 8, 11, 14-15, 17-18, 70-21, 23-25, 35, 37, 39, 48, 58-59, 66, 74, 79-81, 83-84, 86, 88-89,93, 102, 105, 108, 111, 113, 122,146-150, 152-154, 157, 164-167,169-170, 172-175, 177, 181-185,202, 206, 220-222, 230-231, 234-235, 244-245, 254-255, 258-259,262-263, 266-267, 269-270

intended, 14, 22, 24, 87, 102, 277, 284lexical, 155literal, 3, 5, 6, 9, 11, 22-25, 28-30, 34-

35, 38-39, 41, 48, 57-58, 62, 64-66,74, 79-80, 84, 86-89, 95-96, 102-105, 107-108, 114, 125, 129, 135,146-147, 157, 164-168, 170-174,177, 180-183, 203-208, 210, 214-215, 220-222, 230-231, 233-234,244-245, 250, 258-259, 262, 264-265, 268, 276, 283, 314

metaphorical, 30, 32, 84, 94, 245non-compositional, 31, 48, 61sentence, 3, 28, 201, 262-263stipulated, 5, 14, 17, 19, 23-25, 39, 57, 64word, 6, 9, 10, 11, 17, 25, 36, 202, 220

Memory, 4, 6, 9, 11, 30, 42, 130, 136, 146,172, 212, 217-218, 220-221, 239,304, 319

lexical, 33, 41semantic, 42

Metaphor, 18, 22, 29-32, 36, 38, 42, 44, 57,59-60, 66-67, 74, 83, 86, 103-104,120-123, 145, 217, 231-233, 245,250, 275, 277, 280, 289, 292, 300

comprehension, see Comprehensionconceptual, 24, 30, 36, 59-61, 66-73, 75,

231-232conventional, 35, 58, 60, 62, 275, 277,

281, 284dead, 32-33, 36, 57-59, 67, 69, 71, 73-

74, 245, 250Metonymy, 29Modularity, 201-202Motivation, 35-36, 45, 48

N

Need,communicative, 121, 123

Negation, 16, 19

P

Paraphrase,literal, 6, 11, 12, 57

Parser, 82, 85-86, 204, 206, 216, 218-219,223, 236, 238-242, 245, 262

syntactic, 85-86Parsing, 215, 223, 237-238, 240

idiom, 210, 218, 221syntactic, 82

Partial function, 257-259Pattern, 237

matching, 230, 263Performance, 23, 113, 117, 118, 217, 277Polysemy, 10- I1, 17-18,Pragmatic,

function, 268-269Predictability, 35-36, 146, 166, 171, 183,

255-257Predictable, see PredictabilityPriming, 152, 166-168, 170-172, 183, 209,

216, 222conceptual, 214-215, 220contextual, 260cross modal, 165-166, 183, 206, 214, 244effect, 84, 170, 183, 209-216, 218-219, 221syntactic, 218-219

Processing, 49, 57, 75, 91, 93, 102-106, 110,118, 120, 136, 140, 146-148, 155,157, 165, 182, 203-204, 207, 213,215-222, 229-230, 234, 239, 241-243, 245, 250, 263, 287

discourse, 38idiom, 6, 64-65, 73-74, 86, 93, 97, 105,

129, 134-137, 140, 146, 148, 152-156, 165, 166, 202-204, 208, 212,

219, 222, 229, 237-238, 242, 244,277

lexical, 5, 147, 152, 154-155, 216linguistic, 4, 5, 6, 7, 9, 12, 16, 107, 111,

114-115, 222, 277literal, 203-205, 221semantic, 5, 17, 38, 106, 110-111, 121,

203-204, 216

336 SUBJECT INDEX

Processing, (cont.)sentence, 205-207, 267syntactic, 5, 17, 89, 202-204, 206, 208-

210, 213, 215-216, 219Processor, 201, 240-241

language, 166, 171, 173-175, 177-178,182, 185

lexical, 82, 85-86, 201linguistic, 82parsing, 239semantic, 202, 204-206, 214-215syntactic, 85, 201-202, 204, 206, 209-210.

214-215Production, 43, 102-103, 112, 113, 116-117,

134, 145, 216, 277Productivity, 245, 287

discourse, 17, 19, 20, 22, 233-234idiom, 8, 14, 16, 22, 63language, 36lexical, 233semantic, 8, 15-16, 19, 32, 37, 63, 233-

234syntactic, 63, 244

Proverb, 3, 23-24, 27, 33, 103, 121, 157,249

Pseudo- idioms, 130

Q

Quantifier, 272-273, 281, 285, 290, 292-293

R

Recognition, 154-155, 171, 244idiom, 49, 146, 237-238, 241-242, 244,

251, 253, 256, 265-266pattern, 237point, 81, 90, 154-155, 243

Reference, 181, 245, 273, 300Referent, 8, 18, 22, 30, 62, 63, 66, 110-Ill,

120-121, 132, 173, 180-182, 234-236, 251, 276, 280-293

conceptual, 232figurative, 36, 39, 62-63, 66, 234idiomatic, 13-14, 35, 38, 48, 62-63

Relationship,metaphorical, 276, 279

Representation, 201-202, 206, 235-236, 239,242-244, 254, 261-262, 273, 310

discourse, 201, 221-222

idiom, 33, 48, 82-83, 140, 154, 212, 218-220, 232, 236, 271, 284

idiomatic, 91, 222, 242lexical, 12, 85, 135, 146, 164- 165, 203,

217-218, 255, 275literal, 91meaning, 49, 85, 231-232, 235, 260, 266mental, 118, 147, 152, 171, 212semantic, 35, 64, 104, 106, 116, 184, 235,

236, 251, 274surface, 235-236syntactic, 213, 251

Restriction,subcategorization, 129

Retrieval, 41-42, 57, 64, 81, 84, 88, 134-135, 146, 164-167, 172, 203-206,209, 212, 214-215, 217-223, 236,267

Rule,discourse, 14grammatical, 271lexical, 254selectional restriction, 65transformational, 255

S

Semantic,analyser, 86analysis, see Analysisautonomy, 203, 206-208description, 49dominance, 203, 206-208, 210, 214-215field, 41, 63, 66, 106, 114, 124, 232operation, 273processor, see Processorprocessing, see Processingproductivity, see Productivity

Sentence, 6, 11, 21, 30meaning, see Meaningprocessing, see Processing

Significant, 121 -123Speech act, 27

indirect, 28-29, 42, 102-103, 217, 229Strategy, 24, 40, 48

communicative, 18comprehension, 24interpretative, 45, 47-49

Structure,idiom, 48-49semantic, 47, 49, 206, 250

SUBJECT INDEX 337

syntactic, 49, 82, 85, 92, 133, 145, 149,152, 202, 204, 207, 212, 214-218,220, 222, 237, 250, 254

word, 32Syntactic,

analysis, see Analysisdominance, 203-208, 210, 214level, 21operation, 6, 19-22, 80, 82, 145, 184,

271-273, 293parsing, see Parsingprocessing, see Processingsematic dependencies, 202-204, 207structure, see Structurevariation, 254, 268-269variability, 122, 251

T

Theory,grammatical, 250-251, 254, 262semantic, 257

U

Uniqueness point, 81-82, 86-88, 91-93,154-155

Unitlexical, 33, 49, 79, 83, 85, 87, 89, 96, 217,

272semantic, 278syntactic, 292

W

Word, 6, 7, 22, 29-31, 49activation, see Activationconstituent, 3, 5, 9, 11, 12, 13, 27-28, 30,

32-33, 35, 37-38, 48formation, 304-305identification, see Identificationlong, 5, 6, 7, 10meaning, see Meaningsemantic, 7structure, see Structure

Documents

Idioms Processing, Structure and Interpretation - Cristina Cacciari